Department of Informatics

Reinforcement learning (RL) has shown great effectiveness in quadrotor control, enabling specialized policies to develop even human-champion-level performance in single-task scenarios. However, these specialized policies often struggle with novel tasks, requiring a complete retraining of the policy from scratch. To address this limitation, this paper presents a novel multi-task reinforcement learning (MTRL) framework tailored for quadrotor control, leveraging the shared physical dynamics of the platform to enhance sample efficiency and task performance. By employing a multi-critic architecture and shared task encoders, our framework facilitates knowledge transfer across tasks, enabling a single policy to execute diverse maneuvers, including high-speed stabilization, velocity tracking, and autonomous racing. Our experimental results, validated both in simulation and real-world scenarios, demonstrate that our framework outperforms baseline approaches in terms of sample efficiency and overall task performance.

Imitation learning with a privileged teacher has proven effective for learning complex control behaviors from high-dimensional inputs, such as images. In this framework, a teacher is trained with privileged task information, while a student tries to predict the actions of the teacher with more limited observations, e.g., in a robot navigation task, the teacher might have access to distances to nearby obstacles, while the student only receives visual observations of the scene. However, privileged imitation learning faces a key challenge: the student might be unable to imitate the teacher's behavior due to partial observability. This problem arises because the teacher is trained without considering if the student is capable of imitating the learned behavior. To address this teacher-student asymmetry, we propose a framework for joint training of the teacher and student policies, encouraging the teacher to learn behaviors that can be imitated by the student despite the latters limited access to information and its partial observability. Based on the performance bound in imitation learning, we add (i) the approximated action difference between teacher and student as a penalty term to the reward function of the teacher, and (ii) a supervised teacher-student alignment step. We motivate our method with a maze navigation task and demonstrate its effectiveness on complex vision-based quadrotor flight and manipulation tasks.

Is it possible to combine the benefits of model-free reinforcement learning (RL) - known for its strong task performance and flexibility in optimizing general reward formulations - with the robustness and online replanning capabilities of model predictive control (MPC)? This extension digs deeper into the answer by studying our new framework called Actor-Critic Model Predictive Control. We conduct a deep study that exposes the benefits of the proposed approach: it achieves better out-of-distribution behaviour, better robustness to changes in the dynamics and improved sample efficiency. Additionally, we conduct an empirical analysis that reveals a relationship between the critic's learned value function and the cost function of the differentiable MPC, providing a deeper understanding of the interplay between the critic's value and the MPC cost functions. Our method achieves the same superhuman performance as state-of-the-art model-free RL, showcasing speeds of up to 21 m/s.

Visual Odometry (VO) is essential to downstream mobile robotics and augmented/virtual reality tasks. Despite recent advances, existing VO methods still rely on heuristic design choices that require several weeks of hyperparameter tuning by human experts, hindering generalizability and robustness. We address these challenges by reframing VO as a sequential decision-making task and applying Reinforcement Learning (RL) to adapt the VO process dynamically. Our approach introduces a neural network, operating as an agent within the VO pipeline, to make decisions such as keyframe and grid-size selection based on real-time conditions. Our method minimizes reliance on heuristic choices using a reward function based on pose error, runtime, and other metrics to guide the system. Our RL framework treats the VO system and the image sequence as an environment, with the agent receiving observations from keypoints, map statistics, and prior poses. Experimental results using classical VO methods and public benchmarks demonstrate improvements in accuracy and robustness, validating the generalizability of our RL-enhanced VO approach to different scenarios. We believe this paradigm shift advances VO technology by eliminating the need for time-intensive parameter tuning of heuristics.

We present the first vision-based quadrotor system that autonomously navigates through a sequence of gates at high speeds while directly mapping pixels to control commands. Like professional drone-racing pilots, our system does not use explicit state estimation and leverages the same control commands humans use (collective thrust and body rates). We demonstrate agile flight at speeds up to 40km/h with accelerations up to 2g. This is achieved by training vision-based policies with reinforcement learning (RL). The training is facilitated using an asymmetric actor-critic with access to privileged information. To overcome the computational complexity during image-based RL training, we use the inner edges of the gates as a sensor abstraction. Our approach enables autonomous agile flight with standard, off-the-shelf hardware.

While most recent advancements in legged robot control have been driven by model-free reinforcement learning, we explore the potential of differentiable simulation. Differentiable simulation promises faster convergence and more stable training by computing low-variant first-order gradients using the robot model, but so far, its use for legged robot control has remained limited to simulation. The main challenge with differentiable simulation lies in the complex optimization landscape of robotic tasks due to discontinuities in contact-rich environments, e.g., quadruped locomotion. This work proposes a new, differentiable simulation framework to overcome these challenges. The key idea involves decoupling the complex whole-body simulation, which may exhibit discontinuities due to contact, into two separate continuous domains. Subsequently, we align the robot state resulting from the simplified model with a more precise, non-differentiable simulator to maintain sufficient simulation accuracy. Our framework enables learning quadruped walking in minutes using a single simulated robot without any parallelization. When augmented with GPU parallelization, our approach allows the quadruped robot to master diverse locomotion skills, including trot, pace, bound, and gallop, on challenging terrains in minutes. Additionally, our policy achieves robust locomotion performance in the real world zero-shot. To the best of our knowledge, this work represents the first demonstration of using differentiable simulation for controlling a real quadruped robot. This work provides several important insights into using differentiable simulations for legged locomotion in the real world.

We combine the effectiveness of Reinforcement Learning (RL) and the efficiency of Imitation Learning (IL) in the context of vision-based, autonomous drone racing. We focus on directly processing visual input without explicit state estimation. While RL offers a general framework for learning complex controllers through trial and error, it faces challenges regarding sample efficiency and computational demands due to the high dimensionality of visual inputs. Conversely, IL demonstrates efficiency in learning from visual demonstrations but is limited by the quality of those demonstrations and faces issues like covariate shift. To overcome these limitations, we propose a novel training framework combining RL and IL advantages. Our framework involves three stages: (i) initial training of a teacher policy using privileged state information, (ii) distilling this policy into a student policy using IL, (iii) performance-constrained adaptive RL fine-tuning. Our experiments in both simulated and real-world environments demonstrate that our approach achieves superior performance and robustness than IL or RL alone in navigating a quadrotor through a racing course using only visual information without explicit state estimation.

An open research question in robotics is how to combine the benefits of model-free reinforcement learning (RL) - known for its strong task performance and flexibility in optimizing general reward formulations - with the robustness and online replanning capabilities of model predictive control (MPC). This paper provides an answer by introducing a new framework called Actor-Critic Model Predictive Control. The key idea is to embed a differentiable MPC within an actor-critic RL framework. The proposed approach leverages the short-term predictive optimization capabilities of MPC with the exploratory and end-to-end training properties of RL. The resulting policy effectively manages both short-term decisions through the MPC-based actor and long-term prediction via the critic network, unifying the benefits of both model-based control and end-to-end learning. We validate our method in both simulation and the real world with a quadcopter platform across various high-level tasks. We show that the proposed architecture can achieve real-time control performance, learn complex behaviors via trial and error, and retain the robustness inherent to MPC.

Scene transfer for vision-based mobile robotics applications is a highly relevant and challenging problem. The utility of a robot greatly depends on its ability to perform a task in the real world, outside of a well-controlled lab environment. Existing scene transfer end-to-end policy learning approaches often suffer from poor sample efficiency or limited generalization capabilities, making them unsuitable for mobile robotics applications. This work proposes an adaptive multi- pair contrastive learning strategy for visual representation learning that enables zero-shot scene transfer and real-world deployment. Control policies relying on the embedding are able to operate in unseen environments without the need for finetuning in the deployment environment. We demonstrate the performance of our approach on the task of agile, vision-based quadrotor flight. Extensive simulation and real-world experi- ments demonstrate that our approach successfully generalizes beyond the training domain and outperforms all baselines.

In Reinforcement Learning, the trade-off between exploration and exploitation poses a complex challenge for achieving efficient learning from limited samples. While recent works have been effective in leveraging past experiences for policy updates, they often overlook the potential of reusing past experiences for data collection. Independent of the underlying RL algorithm, we introduce the concept of a Contrastive Initial State Buffer, which strategically selects states from past experiences and uses them to initialize the agent in the environment in order to guide it toward more informative states. We validate our approach on two complex robotic tasks without relying on any prior information about the environment: (i) locomotion of a quadruped robot traversing challenging terrains and (ii) a quadcopter drone racing through a track. The experimental results show that our initial state buffer achieves higher task performance than the nominal baseline while also speeding up training convergence.

Why can ReinforcementLearning (RL) achieve results beyond OptimalControl (OC) in many real-world robotics control tasks? We investigate this question in our paper published today in Science Robotics. We argue that this question can be investigated along two axes: the optimization method and the optimization objective. Our results indicate that RL does not outperform OC because RL optimizes its objective better. Rather, RL outperforms OC because it optimizes a better objective. RL can directly optimize a task-level objective and can leverage domain randomization to cope with model uncertainty, allowing the discovery of more robust control responses. Beyond the fundamental study, our work contributes an RL-based controller that delivers the highest performance ever demonstrated on an autonomous racing drone. Our drone achieved remarkable performance: peak acceleration greater than 12 g and peak velocity ~108 km/h, all within minutes of training with a standard workstation.

First-person view (FPV) drone racing is a televised sport in which professional competitors pilot high-speed aircraft through a three-dimensional circuit. Each pilot sees the environment from their drone's perspective via video streamed from an onboard camera. Reaching the level of professional pilots with an autonomous drone is challenging since the robot needs to fly at its physical limits while estimating its speed and location in the circuit exclusively from onboard sensors. Here we introduce Swift, an autonomous system that can race physical vehicles at the level of the human world champions. The system combines deep reinforcement learning in simulation with data collected in the physical world. Swift competed against three human champions, including the world champions of two international leagues, in real-world head-to-head races. Swift won multiple races against each of the human champions and demonstrated the fastest recorded race time. This work represents a milestone for mobile robotics and machine intelligence, which may inspire the deployment of hybrid learning-based solutions in other physical systems.

Model Predictive Control (MPC) has become a popular framework in embedded control for high-performance autonomous systems. However, to achieve good control performance using MPC, an accurate dynamics model is key. To maintain real-time operation, the dynamics models used on embedded systems have been limited to simple first-principle models, which substantially limits their representative power. In contrast to such simple models, machine learning approaches, specifically neural networks, have been shown to accurately model even complex dynamic effects, but their large computational complexity hindered combination with fast real-time iteration loops. With this work, we present Real-time Neural MPC, a framework to efficiently integrate large, complex neural network architectures as dynamics models within a model-predictive control pipeline. Our experiments, performed in simulation and the real world onboard a highly agile quadrotor platform, demonstrate the capabilities of the described system to run learned models with, previously infeasible, large modeling capacity using gradient-based online optimization MPC. Compared to prior implementations of neural networks in online optimization MPC we can leverage models of over 4000 times larger parametric capacity in a 50Hz real-time window on an embedded platform. Further, we show the feasibility of our framework on real-world problems by reducing the positional tracking error by up to 82% when compared to state-of-the-art MPC approaches without neural network dynamics.

Autonomous drones can operate in remote and unstructured environments, enabling various real-world applications. However, the lack of effective vision-based algorithms has been a stumbling block to achieving this goal. Existing systems often require hand-engineered components for state estimation, planning, and control. Such a sequential design involves laborious tuning, human heuristics, and compounding delays and errors. This paper tackles the vision-based autonomous-drone racing problem by learning deep sensorimotor policies. We use contrastive learning to extract robust feature representations from the input images and leverage a two-stage learning-by-cheating framework for training a neural network policy. The resulting policy directly infers control commands with feature representations learned from raw images, forgoing the need for globally-consistent state estimation, trajectory planning, and handcrafted control design. Our experimental results indicate that our vision-based policy can achieve the same level of racing performance as the state-based policy while being robust against different visual disturbances and distractors. This work serves as a stepping-stone toward developing intelligent vision-based autonomous systems that control the drone purely from image inputs, like human pilots.

Control design for robotic systems is complex and often requires solving an optimization to follow a trajectory accurately. Online optimization approaches like Model Predictive Control (MPC) have been shown to achieve great tracking performance, but require high computing power. Conversely, learning-based offline optimization approaches, such as Reinforcement Learning (RL), allow fast and efficient execution on the robot but hardly match the accuracy of MPC in trajectory tracking tasks. In systems with limited compute, such as aerial vehicles, an accurate controller that is efficient at execution time is imperative. We propose an Analytic Policy Gradient (APG) method to tackle this problem. APG exploits the availability of differentiable simulators by training a controller offline with gradient descent on the tracking error. We address training instabilities that frequently occur with APG through curriculum learning and experiment on a widely used controls benchmark, the CartPole, and two common aerial robots, a quadrotor and a fixed-wing drone. Our proposed method outperforms both model-based and model-free RL methods in terms of tracking error. Concurrently, it achieves similar performance to MPC while requiring more than an order of magnitude less computation time. Our work provides insights into the potential of APG as a promising control method for robotics. To facilitate the exploration of APG, we open-source our code and make it publicly available.

Because of their high temporal resolution, increased resilience to motion blur, and very sparse output, event cameras have been shown to be ideal for low-latency and low-bandwidth feature tracking, even in challenging scenarios. Existing feature tracking methods for event cameras are either handcrafted or derived from first principles but require extensive parameter tuning, are sensitive to noise, and do not generalize to different scenarios due to unmodeled effects. To tackle these deficiencies, we introduce the first data-driven feature tracker for event cameras, which leverages low-latency events to track features detected in a grayscale frame. We achieve robust performance via a novel frame attention module, which shares information across feature tracks. By directly transferring zero-shot from synthetic to real data, our data-driven tracker outperforms existing approaches in relative feature age by up to 120 % while also achieving the lowest latency. This performance gap is further increased to 130 % by adapting our tracker to real data with a novel self-supervision strategy.

Recently, learning-based controllers have been shown to push mobile robotic systems to their limits and provide the robustness needed for many real-world applications. However, only classical optimization-based control frameworks offer the inherent flexibility to be dynamically adjusted during execution by, for example, setting target speeds or actuator limits. We present a framework to overcome this shortcoming of neural controllers by conditioning them on an auxiliary input. This advance is enabled by including a feature-wise linear modulation layer (FiLM). We use model-free reinforcement-learning to train quadrotor control policies for the task of navigating through a sequence of waypoints in minimum time. By conditioning the policy on the maximum available thrust or the viewing direction relative to the next waypoint, a user can regulate the aggressiveness of the quadrotor’s flight during deployment. We demonstrate in simulation and in real-world experiments that a single control policy can achieve close to time-optimal flight performance across the entire performance envelope of the robot, reaching up to 60 km/h and 4.5 g in acceleration. The ability to guide a learned controller during task execution has implications beyond agile quadrotor flight, as conditioning the control policy on human intent helps safely bringing learning based systems out of the well-defined laboratory environment into the wild.

Recently, neural control policies have outperformed existing model-based planning-and-control methods for autonomously navigating quadrotors through cluttered environments in minimum time. However, they are not perception aware, a crucial requirement in vision-based navigation due to the camera's limited field of view and the underactuated nature of a quadrotor. We propose a method to learn neural network policies that achieve perception-aware, minimum-time flight in cluttered environments. Our method combines imitation learning and reinforcement learning (RL) by leveraging a privileged learning-by-cheating framework. Using RL, we first train a perception-aware teacher policy with full-state information to fly in minimum time through cluttered environments. Then, we use imitation learning to distill its knowledge into a vision-based student policy that only perceives the environment via a camera. Our approach tightly couples perception and control, showing a significant advantage in computation speed (10x faster) and success rate. We demonstrate the closed-loop control performance using a physical quadrotor and hardware-in-the-loop simulation at speeds up to 50 km/h.

Event cameras are bio-inspired vision sensors that naturally capture the dynamics of a scene, filtering out redundant information. This paper presents a deep neural network approach that unlocks the potential of event cameras on a challenging motion-estimation task: prediction of a vehicle's steering angle. To make the best out of this sensor-algorithm combination, we adapt state-of-the-art convolutional architectures to the output of event sensors and extensively evaluate the performance of our approach on a publicly available large scale event-camera dataset (~1000 km). We present qualitative and quantitative explanations of why event cameras allow robust steering prediction even in cases where traditional cameras fail, e.g. challenging illumination conditions and fast motion. Finally, we demonstrate the advantages of leveraging transfer learning from traditional to event-based vision, and show that our approach outperforms state-of-the-art algorithms based on standard cameras.

Retrieving accurate semantic information in challenging high dynamic range (HDR) and high-speed conditions remains an open challenge for image-based algorithms due to severe image degradations. Event cameras promise to address these challenges since they feature a much higher dynamic range and are resilient to motion blur. Nonetheless, semantic segmentation with event cameras is still in its infancy which is chiefly due to the lack of high-quality, labeled datasets. In this work, we introduce ESS (Event-based Semantic Segmentation), which tackles this problem by directly transferring the semantic segmentation task from existing labeled image datasets to unlabeled events via unsupervised domain adaptation (UDA). Compared to existing UDA methods, our approach aligns recurrent, motion-invariant event embeddings with image embeddings. For this reason, our method neither requires video data nor per-pixel alignment between images and events and, crucially, does not need to hallucinate motion from still images. Additionally, we introduce DSEC-Semantic, the first large-scale event-based dataset with fine-grained labels. We show that using image labels alone, ESS outperforms existing UDA approaches, and when combined with event labels, it even outperforms state-of-the-art supervised approaches on both DDD17 and DSEC-Semantic. Finally, ESS is general-purpose, which unlocks the vast amount of existing labeled image datasets and paves the way for new and exciting research directions in new fields previously inaccessible for event cameras.

Modern high dynamic range (HDR) imaging pipelines align and fuse multiple low dynamic range (LDR) images captured at different exposure times. While these methods work well in static scenes, dynamic scenes remain a challenge since the LDR images still suffer from saturation and noise. In such scenarios, event cameras would be a valid complement, thanks to their higher temporal resolution and dynamic range. In this paper, we propose the first multi- bracket HDR pipeline combining a standard camera with an event camera. Our results show better overall robustness when using events, with improvements in PSNR by up to 5dB on synthetic data and up to 0.7dB on real-world data. We also introduce a new dataset containing bracketed LDR images with aligned events and HDR ground truth.

Humans race drones faster than neural networks trained for end-to-end autonomous flight. This may be related to the ability of human pilots to select task-relevant visual information effectively. This work investigates whether neural networks capable of imitating human eye gaze behavior and attention can improve neural network performance for the challenging task of vision-based autonomous drone racing. We hypothesize that gaze-based attention prediction can be an efficient mechanism for visual information selection and decision making in a simulator-based drone racing task. We test this hypothesis using eye gaze and flight trajectory data from 18 human drone pilots to train a visual attention prediction model. We then use this visual attention prediction model to train an end-to-end controller for vision-based autonomous drone racing using imitation learning. We compare the drone racing performance of the attention-prediction controller to those using raw image inputs and image-based abstractions (i.e., feature tracks). Comparing success rates for completing a challenging race track by autonomous flight, our results show that the attention-prediction based controller (88% success rate) outperforms the RGB-image (61% success rate) and feature-tracks (55% success rate) controller baselines. Furthermore, visual attention-prediction and feature-track based models showed better generalization performance than image-based models when evaluated on hold-out reference trajectories. Our results demonstrate that human visual attention prediction improves the performance of autonomous vision-based drone racing agents and provides an essential step towards vision-based, fast, and agile autonomous flight that eventually can reach and even exceed human performances.

Quadrotors are highly nonlinear dynamical systems that require carefully tuned controllers to be pushed to their physical limits. Recently, learning-based control policies have been proposed for quadrotors, as they would potentially allow learning direct mappings from high-dimensional raw sensory observations to actions. Due to sample inefficiency, training such learned controllers on the real platform is impractical or even impossible. Training in simulation is attractive but requires to transfer policies between domains, which demands trained policies to be robust to such domain gap. In this work, we make two contributions: (i) we perform the first benchmark comparison of existing learned control policies for agile quadrotor flight and show that training a control policy that commands body-rates and thrust results in more robust sim-to-real transfer compared to a policy that directly specifies individual rotor thrusts, (ii) we demonstrate for the first time that such a control policy trained via deep reinforcement learning can control a quadrotor in real-world experiments at speeds over 45km/h.

Event cameras are novel sensors with outstanding properties such as high temporal resolution and high dynamic range. Despite these characteristics, event-based vision has been held back by the shortage of labeled datasets due to the novelty of event cameras. To overcome this drawback, we propose a task transfer method that allows models to be trained directly with labeled images and unlabeled event data. Compared to previous approaches, (i) our method transfers from single images to events instead of high frame rate videos, and (ii) does not rely on paired sensor data. To achieve this, we leverage the generative event model to split event features into content and motion features. This feature split enables to efficiently match the latent space for events and images, which is crucial for a successful task transfer. Thus, our approach unlocks the vast amount of existing image datasets for the training of event-based neural networks. Our task transfer method consistently outperforms methods applicable in the Unsupervised Domain Adaptation setting for object detection by 0.26 mAP (increase by 93%) and classification by 2.7% accuracy.

Policy Search and Model Predictive Control (MPC) are two different paradigms for robot control: policy search has the strength of automatically learning complex policies using experienced data, while MPC can offer optimal control performance using models and trajectory optimization. An open research question is how to leverage and combine the advantages of both approaches. In this work, we provide an answer by using policy search for automatically choosing high-level decision variables for MPC, which leads to a novel policy-search-for-model-predictive-control framework. Specifically, we formulate the MPC as a parameterized controller, where the hard-to-optimize decision variables are represented as high-level policies. Such a formulation allows optimizing policies in a self-supervised fashion. We validate this framework by focusing on a challenging problem in agile drone flight: flying a quadrotor through fast-moving gates. Experiments show that our controller achieves robust and real-time control performance in both simulation and the real world. The proposed framework offers a new perspective for merging learning and control.

Quadrotors are agile. Unlike most other machines, they can traverse extremely complex environments at high speeds. To date, only expert human pilots have been able to fully exploit their capabilities. Autonomous operation with onboard sensing and computation has been limited to low speeds. State-of-the-art methods generally separate the navigation problem into subtasks: sensing, mapping, and planning. While this approach has proven successful at low speeds, the separation it builds upon can be problematic for high-speed navigation in cluttered environments. Indeed, the subtasks are executed sequentially, leading to increased processing latency and a compounding of errors through the pipeline. Here we propose an end-to-end approach that can autonomously fly quadrotors through complex natural and man-made environments at high speeds, with purely onboard sensing and computation. The key principle is to directly map noisy sensory observations to collision-free trajectories in a receding-horizon fashion. This direct mapping drastically reduces processing latency and increases robustness to noisy and incomplete perception. The sensorimotor mapping is performed by a convolutional network that is trained exclusively in simulation via privileged learning: imitating an expert with access to privileged information. By simulating realistic sensor noise, our approach achieves zero-shot transfer from simulation to challenging real-world environments that were never experienced during training: dense forests, snow-covered terrain, derailed trains, and collapsed buildings. Our work demonstrates that end-to-end policies trained in simulation enable high-speed autonomous flight through challenging environments, outperforming traditional obstacle avoidance pipelines. We release the code open source.

Quadrotors are extremely agile, so much in fact, that classic first-principle-models come to their limits. Aerodynamic effects, while insignificant at low speeds, become the dominant model defect during high speeds or agile maneuvers. Accurate modeling is needed to design robust high-performance control systems and enable flying close to the platform's physical limits. We propose a hybrid approach fusing first principles and learning to model quadrotors and their aerodynamic effects with unprecedented accuracy. First principles fail to capture such aerodynamic effects, rendering traditional approaches inaccurate when used for simulation or controller tuning. Data-driven approaches try to capture aerodynamic effects with blackbox modeling, such as neural networks; however, they struggle to robustly generalize to arbitrary flight conditions. Our hybrid approach unifies and outperforms both first-principles blade-element theory and learned residual dynamics. It is evaluated in one of the world's largest motion-capture systems, using autonomous-quadrotor-flight data at speeds up to 65km/h. The resulting model captures the aerodynamic thrust, torques, and parasitic effects with astonishing accuracy, outperforming existing models with 50% reduced prediction errors, and shows strong generalization capabilities beyond the training set.

Professional race-car drivers can execute extreme overtaking maneuvers. However, existing algorithms for autonomous overtaking either rely on simplified assumptions about the vehicle dynamics or try to solve expensive trajectory-optimization problems online. When the vehicle approaches its physical limits, existing model-based controllers struggle to handle highly nonlinear dynamics, and cannot leverage the large volume of data generated by simulation or real-world driving. To circumvent these limitations, we propose a new learning-based method to tackle the autonomous overtaking problem. We evaluate our approach in the popular car racing game Gran Turismo Sport, which is known for its detailed modeling of various cars and tracks. By leveraging curriculum learning, our approach leads to faster convergence as well as increased performance compared to vanilla reinforcement learning. As a result, the trained controller outperforms the built-in model-based game AI and achieves comparable overtaking performance with an experienced human driver.

Once an academic venture, autonomous driving has received unparalleled corporate funding in the last decade. Still, operating conditions of current autonomous cars are mostly restricted to ideal scenarios. This means that driving in challenging illumination conditions such as night, sunrise, and sunset remains an open problem. In these cases, standard cameras are being pushed to their limits in terms of low light and high dynamic range performance. To address these challenges, we propose, DSEC, a new dataset that contains such demanding illumination conditions and provides a rich set of sensory data. DSEC offers data from a wide-baseline stereo setup of two color frame cameras and two high-resolution monochrome event cameras. In addition, we collect lidar data and RTK GPS measurements, both hardware synchronized with all camera data. One of the distinctive features of this dataset is the inclusion of high-resolution event cameras. Event cameras have received increasing attention for their high temporal resolution and high dynamic range performance. However, due to their novelty, event camera datasets in driving scenarios are rare. This work presents the first high resolution, large scale stereo dataset with event cameras. The dataset contains 53 sequences collected by driving in a variety of illumination conditions and provides ground truth disparity for the development and evaluation of event-based stereo algorithms.

In many robotic tasks, such as drone racing, the goal is to travel through a set of waypoints as fast as possible. A key challenge for this task is planning the minimum-time trajectory, which is typically solved by assuming perfect knowledge of the waypoints to pass in advance. The resulting solutions are either highly specialized for a single-track layout, or suboptimal due to simplifying assumptions about the platform dynamics. In this work, a new approach to minimum-time trajectory generation for quadrotors is presented. Leveraging deep reinforcement learning and relative gate observations, this approach can adaptively compute near-time-optimal trajectories for random track layouts. Our method exhibits a significant computational advantage over approaches based on trajectory optimization for non-trivial track configurations. The proposed approach is evaluated on a set of race tracks in simulation and the real world, achieving speeds of up to 17 m/s with a physical quadrotor.

Due to noisy actuation and external disturbances, tuning controllers for high-speed flight is very challenging. In this paper, we ask the following questions: How sensitive are controllers to tuning when tracking high-speed maneuvers? What algorithms can we use to automatically tune them? To answer the first question, we study the relationship between parameters and performance and find out that the faster the maneuver, the more sensitive a controller becomes to its parameters. To answer the second question, we review existing methods for controller tuning and discover that prior works often perform poorly on the task of high-speed flight. Therefore, we propose AutoTune, a sampling-based tuning algorithm specifically tailored to high-speed flight. In contrast to previous work, our algorithm does not assume any prior knowledge of the drone or its optimization function and can deal with the multi-modal characteristics of the parameters' optimization space. We thoroughly evaluate AutoTune both in simulation and in the physical world. In our experiments, we outperform existing tuning algorithms by up to 90\% in trajectory completion. The resulting controllers are tested in the AirSim Game of Drones competition, where we outperform the winner by up to 25\% in lap-time. Finally, we show that AutoTune improves tracking error when flying a physical platform with respect to parameters tuned by a human expert.

Recent works in geometric deep learning have introduced neural networks that allow performing inference tasks on three-dimensional geometric data by defining convolution, and sometimes pooling, operations on triangle meshes. These methods, however, either consider the input mesh as a graph, and do not exploit specific geometric properties of meshes for feature aggregation and downsampling, or are specialized for meshes, but rely on a rigid definition of convolution that does not properly capture the local topology of the mesh. We propose a method that combines the advantages of both types of approaches, while addressing their limitations: we extend a primal-dual framework drawn from the graph-neural-network literature to triangle meshes, and define convolutions on two types of graphs constructed from an input mesh. Our method takes features for both edges and faces of a 3D mesh as input and dynamically aggregates them using an attention mechanism. At the same time, we introduce a pooling operation with a precise geometric interpretation, that allows handling variations in the mesh connectivity by clustering mesh faces in a task-driven fashion. We provide theoretical insights of our approach using tools from the mesh-simplification literature. In addition, we validate experimentally our method in the tasks of shape classification and shape segmentation, where we obtain comparable or superior performance to the state of the art.

Event cameras are novel sensors that output brightness changes in the form of a stream of asynchronous "events" instead of intensity frames. Compared to conventional image sensors, they offer significant advantages: high temporal resolution, high dynamic range, no motion blur, and much lower bandwidth. Recently, learning-based approaches have been applied to event-based data, thus unlocking their potential and making significant progress in a variety of tasks, such as monocular depth prediction. Most existing approaches use standard feed-forward architectures to generate network predictions, which do not leverage the temporal consistency presents in the event stream. We propose a recurrent architecture to solve this task and show significant improvement over standard feed-forward methods. In particular, our method generates dense depth predictions using a monocular setup, which has not been shown previously. We pretrain our model using a new dataset containing events and depth maps recorded in the CARLA simulator. We test our method on the Multi Vehicle Stereo Event Camera Dataset (MVSEC). Quantitative experiments show up to 50% improvement in average depth error with respect to previous event-based methods.

Event-based cameras record an asynchronous stream of per-pixel brightness changes. As such, they have numerous advantages over the standard frame-based cameras,including high temporal resolution, high dynamic range, and no motion blur. Due to the asynchronous nature, efficient learning of compact representation for event data is challenging. While it remains not explored the extent to which the spatial and temporal event "information" is useful for pattern recognition tasks. Inthis paper, we focus on single-layer architectures. We analyze the performance of two general problem formulations: the directand the inverse, for unsupervised feature learning from local event data (local volumes of events described in space-time).We identify and show the main advantages of each approach.Theoretically, we analyze guarantees for an optimal solution,possibility for asynchronous, parallel parameter update, and the computational complexity. We present numerical experiments for object recognition. We evaluate the solution under the direct and the inverse problem and give a comparison with the state-of-the-art methods. Our empirical results highlight the advantages of both approaches for representation learning from event data. Weshow improvements of up to 9%in the recognition accuracy compared to the state-of-the-art methods from the same class of methods.

Currently available quadrotor simulators have a rigid and highly-specialized structure: either are they really fast, physically accurate, or photo-realistic. In this work, we propose a paradigm-shift in the development of simulators: moving the trade-off between accuracy and speed from the developers to the end-users. We release a new modular quadrotor simulator: Flightmare. Flightmare is composed of two main components: a configurable rendering engine built on Unity and a flexible physics engine for dynamics simulation. Those two components are totally decoupled and can run independently from each other. Flightmare comes with several desirable features: (i) a large multi-modal sensor suite, including an interface to extract the 3D point-cloud of the scene; (ii) an API for reinforcement learning which can simulate hundreds of quadrotors in parallel; and (iii) an integration with a virtual-reality headset for interaction with the simulated environment. Flightmare can be used for various applications, including path-planning, reinforcement learning, visual-inertial odometry, deep learning, human-robot interaction, etc.

Autonomous car racing raises fundamental robotics challenges such as planning minimum-time trajectories under uncertain dynamics and controlling the car at its friction limits. In this project, we consider the task of autonomous car racing in the top-selling car racing game Gran Turismo Sport. Gran Turismo Sport is known for its detailed physics simulation of various cars and tracks. Our approach makes use of maximum-entropy deep reinforcement learning and a new reward design to train a sensorimotor policy to complete a given race track as fast as possible. We evaluate our approach in three different time trial settings with different cars and tracks. Our results show that the obtained controllers not only beat the built-in non-player character of Gran Turismo Sport, but also outperform the fastest known times in a dataset of personal best lap times of over 50,000 human drivers.

The combination of policy search and deep neural networks holds the promise of automating a variety of decision- making tasks. Model Predictive Control (MPC) provides robust solutions to robot control tasks by making use of a dynamical model of the system and solving an optimization problem online over a short planning horizon. In this work, we leverage probabilistic decision-making approaches and the generalization capability of artificial neural networks to the powerful online optimization by learning a deep high-level policy for the MPC (High-MPC). Conditioning on robot’s local observations, the trained neural network policy is capable of adaptively selecting high-level decision variables for the low-level MPC controller, which then generates optimal control commands for the robot. First, we formulate the search of high-level decision variables for MPC as a policy search problem, specifically, a probabilistic inference problem. The problem can be solved in a closed-form solution. Second, we propose a self-supervised learning algorithm for learning a neural network high-level policy, which is useful for online hyperparameter adaptations in highly dynamic environments. We demonstrate the importance of incorporating the online adaption into autonomous robots by using the proposed method to solve a challenging control problem, where the task is to control a simulated quadrotor to fly through a swinging gate. We show that our approach can handle situations that are difficult for standard MPC.

Event cameras are bio-inspired sensors that respond to per-pixel brightness changes in the form of asynchronous and sparse "events". Recently, pattern recognition algorithms, such as learning-based methods, have made significant progress with event cameras by converting events into synchronous dense, image-like representations and applying traditional machine learning methods developed for standard cameras. However, these approaches discard the spatial and temporal sparsity inherent in event data at the cost of higher computational complexity and latency. In this work, we present a general framework for converting models trained on synchronous image-like event representations into asynchronous models with identical output, thus directly leveraging the intrinsic asynchronous and sparse nature of the event data. We show both theoretically and experimentally that this drastically reduces the computational complexity and latency of high-capacity, synchronous neural networks without sacrificing accuracy. In addition, our framework has several desirable characteristics: (i) it exploits spatio-temporal sparsity of events explicitly, (ii) it is agnostic to the event representation, network architecture, and task, and (iii) it does not require any train-time change, since it is compatible with the standard neural networks' training process. We thoroughly validate the proposed framework on two computer vision tasks: object detection and object recognition. In these tasks, we reduce the computational complexity up to 20 times with respect to high-latency neural networks. At the same time, we outperform state-of-the-art asynchronous approaches up to 24% in prediction accuracy.

Motivated by the astonishing capabilities of natural intelligent agents and inspired by theories from psychology, this paper explores the idea that perception gets coupled to 3D properties of the world via interaction with the environment. Existing works for depth estimation require either massive amounts of annotated training data or some form of hard-coded geometrical constraint. This paper explores a new approach to learning depth perception requiring neither of those. Specifically, we train a specialized global-local network architecture with what would be available to a robot interacting with the environment: from extremely sparse depth measurements down to even a single pixel per image. From a pair of consecutive images, our proposed network outputs a latent representation of the observer’s motion between the images and a dense depth map.

Experiments on several datasets show that, when ground truth is available even for just one of the image pixels, the proposed network can learn monocular dense depth estimation up to 22.5% more accurately than state-of-the-art approaches. We believe that this work, despite its scientific interest, lays the foundations to learn depth from extremely sparse supervision, which can be valuable to all robotic systems acting under severe bandwidth or sensing constraints.

Performing acrobatic maneuvers with quadrotors is extremely challenging. Acrobatic flight requires high thrust and extreme angular accelerations that push the platform to its physical limits. Professional drone pilots often measure their level of mastery by flying such maneuvers in competitions. In this paper, we propose to learn a sensorimotor policy that enables an autonomous quadrotor to fly extreme acrobatic maneuvers with only onboard sensing and computation. We train the policy entirely in simulation by leveraging demonstrations from an optimal controller that has access to privileged information. We use appropriate abstractions of the visual input to enable transfer to a real quadrotor. We show that the resulting policy can be directly deployed in the physical world without any fine-tuning on real data. Our methodology has several favorable properties: it does not require a human expert to provide demonstrations, it cannot harm the physical system during training, and it can be used to learn maneuvers that are challenging even for the best human pilots. Our approach enables a physical quadrotor to fly maneuvers such as the Power Loop, the Barrel Roll, and the Matty Flip, during which it incurs accelerations of up to 3g.

Visual Localization is one of the key enabling technologies for autonomous driving and augmented reality. High quality datasets with accurate 6 Degree-of-Freedom (DoF) reference poses are the foundation for benchmarking and improving existing methods. Traditionally, reference poses have been obtained via Structure-from-Motion (SfM). However, SfM itself relies on local features which are prone to fail when images were taken under different conditions, e.g., day/night changes. At the same time, manually annotating feature correspondences is not scalable and potentially inaccurate. In this work, we propose a semi-automated approach to generate reference poses based on feature matching between renderings of a 3D model and real images via learned features. Given an initial pose estimate, our approach iteratively refines the pose based on feature matches against a rendering of the model from the current pose estimate. We significantly improve the nighttime reference poses of the popular Aachen Day-Night dataset, showing that state-of-the-art visual localization methods perform better (up to 47%) than predicted by the original reference poses. We extend the dataset with new nighttime test images, provide uncertainty estimates for our new reference poses, and introduce a new evaluation criterion. We will make our reference poses and our framework publicly available upon publication.

Spiking Neural Networks (SNNs) are bio-inspired networks that process information conveyed as temporal spikes rather than numeric values. An example of a sensor providing such data is the event camera. It only produces an event when a pixel reports a significant brightness change. Similarly, the spiking neuron of an SNN only produces a spike whenever a significant number of spikes occur within a short period of time. Due to their spike-based computational model, SNNs can process output from event-based, asynchronous sensors without any pre-processing at extremely lower power unlike standard artificial neural networks. This is possible due to specialized neuromorphic hardware that implements the highly-parallelizable concept of SNNs in silicon. Yet, SNNs have not enjoyed the same rise of popularity as artificial neural networks. This not only stems from the fact that their input format is rather unconventional but also due to the challenges in training spiking networks. Despite their temporal nature and recent algorithmic advances, they have been mostly evaluated on classification problems. We propose, for the first time, a temporal regression problem of numerical values given events from an event camera.

We specifically investigate the prediction of the 3-DOF angular velocity of a rotating event camera with an SNN. The difficulty of this problem arises from the prediction of angular velocities continuously in time directly from irregular, asynchronous event-based input. Directly utilising the output of event cameras without any pre-processing ensures that we inherit all the benefits that they provide over conventional cameras. That is high-temporal resolution, high-dynamic range and no motion blur. To assess the performance of SNNs on this task, we introduce a synthetic event camera dataset generated from real-world panoramic images and show that we can successfully train an SNN to perform angular velocity regression.

Neural networks predictions are unreliable when the input sample is out of the training data distribution or corrupted by noise. Being able to detect such failures automatically is fundamental to integrate deep learning algorithms into robotic systems. Current approaches for uncertainty estimation of neural networks require changes to the network and optimization process, typically ignore prior knowledge about the data, and tend to make over-simplifying assumptions which underestimate uncertainty. To address these limitations, we propose a novel framework for uncertainty estimation. Based on Bayesian belief networks and Monte-Carlo sampling, our framework not only fully models the different sources of prediction uncertainty, but also incorporates prior data information, e.g. sensor noise. We show theoretically that this gives us the ability to capture uncertainty better than existing methods. In addition, our framework has several desirable properties: (i) it is agnostic to the network architecture and task; (ii) it does not require changes in the optimization process; (iii) it can be applied to already trained architectures. We thoroughly validate the proposed framework through extensive experiments on both computer vision and control tasks, where we outperform previous methods by up to 23%.

Event cameras are novel sensors that output brightness changes in the form of a stream of asynchronous "events" instead of intensity frames. They offer significant advantages with respect to conventional cameras: high dynamic range (HDR), high temporal resolution, and no motion blur. Recently, novel learning approaches operating on event data have achieved impressive results. Yet, these methods require a large amount of event data for training, which is hardly available due the novelty of event sensors in computer vision research. In this paper, we present a method that addresses these needs by converting any existing video dataset recorded with conventional cameras to \emph{synthetic} event data. This unlocks the use of a virtually unlimited number of existing video datasets for training networks designed for real event data. We evaluate our method on two relevant vision tasks, i.e., object recognition and semantic segmentation, and show that models trained on synthetic events have several benefits: (i) they generalize well to real event data, even in scenarios where standard-camera images are blurry or overexposed, by inheriting the outstanding properties of event cameras; (ii) they can be used for fine-tuning on real data to improve over state-of-the-art for both classification and semantic segmentation.

Dynamically changing environments, unreliable state estimation, and operation under severe resource constraints are fundamental challenges for robotics, which still limit the deployment of small autonomous drones. We address these challenges in the context of autonomous, vision-based drone racing in dynamic environments. A racing drone must traverse a track with possibly moving gates at high speed. We enable this functionality by combining the performance of a state-of-the-art path-planning and control system with the perceptual awareness of a convolutional neural network (CNN). The CNN directly maps raw images to a desired waypoint and speed. Given the CNN output, the planner generates a short minimum-jerk trajectory segment that is tracked by a model-based controller to actuate the drone towards the waypoint. The resulting modular system has several desirable features: (i) it can run fully on-board, (ii) it does not require globally consistent state estimation, and (iii) it is both platform and domain independent. We extensively test the precision and robustness of our system, both in simulation and on a physical platform. In both domains, our method significantly outperforms the prior state of the art. In order to understand the limits of our approach, we additionally compare against professional human drone pilots with different skill levels.

Event cameras are novel sensors that report brightness changes in the form of a stream of asynchronous events instead of intensity frames. They offer significant advantages with respect to conventional cameras: high temporal resolution, high dynamic range, and no motion blur. While the stream of events encodes in principle the complete visual signal, the reconstruction of an intensity image from a stream of events is an ill-posed problem in practice. Existing reconstruction approaches are based on hand-crafted priors and strong assumptions about the imaging process as well as the statistics of natural images.

In this work we propose to learn to reconstruct intensity images from event streams directly from data instead of relying on any hand-crafted priors. We propose a novel recurrent network to reconstruct videos from a stream of events, and train it on a large amount of simulated event data. During training we propose to use a perceptual loss to encourage reconstructions to follow natural image statistics. We further extend our approach to synthesize color images from color event streams.

Our quantitative experiments show that our network surpasses state-of-the-art reconstruction methods by a large margin in terms of image quality (> 20%), while comfortably running in real-time. We show that the network is able to synthesize high framerate videos (> 5,000 frames per second) of high-speed phenomena (e.g. a bullet hitting an object) and is able to provide high dynamic range reconstructions in challenging lighting conditions. As an additional contribution, we demonstrate the effectiveness of our reconstructions as an intermediate representation for event data. We show that off-the-shelf computer vision algorithms can be applied to our reconstructions for tasks such as object classification and visual-inertial odometry and that this strategy consistently outperforms algorithms that were specifically designed for event data. We release the reconstruction code and a pre-trained model to enable further research.

We presented our approach in two different papers (references below). Our first paper (CVPR19) introduced the network architecture (a simple recurrent neural network), the training data, and our first video reconstruction results. In our follow-up paper (T-PAMI), we improved the network architecture by using convolutional LSTM blocks and a temporal consistency loss, leading to improved stability and temporal consistency. Furthermore, the improved network now works well with windows containing variable number of events, which allows to synthesize videos at a very high framerate (> 5,000 frames per second), which we additionally demonstrated in a series of new experiments featuring extremely fast motions.

PULP-DroNet is a deep learning-powered visual navigation engine that enables autonomous navigation of a pocket-size quadrotor in a previously unseen environment. Thanks to PULP-DroNet the nano-drone can explore the environment, avoiding collisions also with dynamic obstacles, in complete autonomy -- no human operator, no ad-hoc external signals, and no remote laptop! This means that all the complex computations are done directly aboard the vehicle and very fast. The visual navigation engine is composed of both a software and a hardware part. The former is based on the previous DroNet project developed by the RPG from the University of Zürich (UZH). DroNet is a shallow convolutional neural network (CNN) which has been used to control a standard-size quadrotor in a set of environments via remote computation. The hardware soul of PULP-DroNet is embodied by the PULP-Shield an ultra-low power visual navigation module featuring a Parallel Ultra-Low-Power (PULP) GAP8 System-on-Chip (SoC) from GreenWaves Technologies (GWT), an ultra-low power camera, and off-chip Flash/DRAM memory; the shield is designed as a pluggable PCB for the Crazyflie 2.0 nano-drone. Then, we developed a general methodology for deploying state-of-the-art deep learning algorithms on top of ultra-low power embedded computation nodes, like a miniaturized drone. Our novel methodology allowed us first to deploy DroNet on the PULP-Shield, and then demonstrating how it enables the execution the CNN on board the CrazyFlie 2.0 within only 64-284mW and with a throughput of 6-18 frame-per-second! Finally, we field-prove our methodology presenting a closed-loop fully working demonstration of vision-driven autonomous navigation relying only on onboard resources, and within an ultra-low power budget. See the videos on the PULP Platform Youtube channel (Video1). We release here, as open source, all our code, hardware designs, datasets, and trained networks.

Event cameras are vision sensors that record asynchronous streams of per-pixel brightness changes, referred to as "events". They have appealing advantages over frame-based cameras for computer vision, including high temporal resolution, high dynamic range, and no motion blur. Due to the sparse, non-uniform spatiotemporal layout of the event signal, pattern recognition algorithms typically aggregate events into a grid-based representation and subsequently process it by a standard vision pipeline, e.g., Convolutional Neural Network (CNN). In this work, we introduce a general framework to convert event streams into grid-based representations through a sequence of differentiable operations. Our framework comes with two main advantages: (i) allows learning the input event representation together with the task dedicated network in an end to end manner, and (ii) lays out a taxonomy that unifies the majority of extant event representations in the literature and identifies novel ones. Empirically, we show that our approach to learning the event representation end-to-end yields an improvement of approximately 12% on optical flow estimation and object recognition over state-of-the-art methods.

Event cameras are novel sensors that report brightness changes in the form of asynchronous “events” instead of intensity frames. They have significant advantages over conventional cameras: high temporal resolution, high dynamic range, and no motion blur. Since the output of event cameras is fundamentally different from conventional cameras, it is commonly accepted that they require the development of specialized algorithms to accommodate the particular nature of events. In this work, we take a different view and propose to apply existing, mature computer vision techniques to videos reconstructed from event data. We propose a novel recurrent network to reconstruct videos from a stream of events, and train it on a large amount of simulated event data. Our experiments show that our approach surpasses state-of-the-art reconstruction methods by a large margin (20%) in terms of image quality. We further apply off-the-shelf computer vision algorithms to videos reconstructed from event data on tasks such as object classification and visual-inertial odometry, and show that this strategy consistently outperforms algorithms that were specifically designed for event data. We believe that our approach opens the door to bringing the outstanding properties of event cameras to an entirely new range of tasks.

Civilian drones are soon expected to be used in a wide variety of tasks, such as aerial surveillance, delivery, or monitoring of existing architectures. Nevertheless, their deployment in urban environments has so far been limited. Indeed, in unstructured and highly dynamic scenarios drones face numerous challenges to navigate autonomously in a feasible and safe way. In contrast to the traditional map-localize-plan methods, this paper explores a data-driven approach to cope with the above challenges. To do this, we propose DroNet, a convolutional neural network that can safely drive a drone through the streets of a city. Designed as a fast 8-layers residual network, DroNet produces, for each single input image, two outputs: a steering angle, to keep the drone navigating while avoiding obstacles, and a collision probability, to let the UAV recognize dangerous situations and promptly react to them. But how to collect enough data in an unstructured outdoor environment, such as a city? Clearly, having an expert pilot providing training trajectories is not an option given the large amount of data required and, above all, the risk that it involves for others vehicles or pedestrians moving in the streets. Therefore, we propose to train a UAV from data collected by cars and bicycles, which, already integrated into urban environments, would expose other cars and pedestrians to no danger. Although trained on city streets, from the viewpoint of urban vehicles, the navigation policy learned by DroNet is highly generalizable. Indeed, it allows a UAV to successfully fly at relative high altitudes, and even in indoor environments, such as parking lots and corridors.

We study the problem of perceiving forest or mountain trails from a single monocular image acquired from the viewpoint of a robot traveling on the trail itself. Previous literature focused on trail segmentation, and used low-level features such as image saliency or appearance contrast; we propose a different approach based on a Deep Neural Network used as a supervised image classifier. By operating on the whole image at once, our system outputs the main direction of the trail compared to the viewing direction. Qualitative and quantitative results computed on a large real-world dataset (which we provide for download) show that our approach outperforms alternatives, and yields an accuracy comparable to the accuracy of humans that are tested on the same image classification task. Preliminary results on using this information for quadrotor control in unseen trails are reported. To the best of our knowledge, this is the first paper that describes an approach to perceive forest trials which is demonstrated on a quadrotor micro aerial vehicle.

We consider the problem of performing rapid training of a terrain classifier in the context of a collaborative robotic search and rescue system. Our system uses a vision-based flying robot to guide a ground robot through unknown terrain to a goal location by building a map of terrain class and elevation. However, due to the unknown environments present in search and rescue scenarios, our system requires a terrain classifier that can be trained and deployed quickly, based on data collected on the spot. We investigate the relationship of training set size and complexity on training time and accuracy, for both feature-based and convolutional neural network classifiers in this scenario. Our goal is to minimize the deployment time of the classi.er in our terrain mapping system within acceptable classi.cation accuracy tolerances. So we are not concerned with training a classifier that generalizes well, only one that works well for this particular environment. We demonstrate that we can launch our aerial robot, gather data, train a classifier, and begin building a terrain map after only 60 seconds of flight.