Abstract:Reinforcement learning (RL) agents need to explore their environment to learn optimal behaviors and achieve maximum rewards. However, exploration can be risky when training RL directly on real systems, while simulation-based training introduces the tricky issue of the sim-to-real gap. Recent approaches have leveraged safety filters, such as control barrier functions (CBFs), to penalize unsafe actions during RL training. However, the strong safety guarantees of CBFs rely on a precise dynamic model. In practice, uncertainties always exist, including internal disturbances from the errors of dynamics and external disturbances such as wind. In this work, we propose a new safe RL framework based on disturbance rejection-guarded learning, which allows for an almost model-free RL with an assumed but not necessarily precise nominal dynamic model. We demonstrate our results on the Safety-gym benchmark for Point and Car robots on all tasks where we can outperform state-of-the-art approaches that use only residual model learning or a disturbance observer (DOB). We further validate the efficacy of our framework using a physical F1/10 racing car. Videos: https://sites.google.com/view/res-dob-cbf-rl
Abstract:Modern non-linear model-based controllers require an accurate physics model and model parameters to be able to control mobile robots at their limits. Also, due to surface slipping at high speeds, the friction parameters may continually change (like tire degradation in autonomous racing), and the controller may need to adapt rapidly. Many works derive a task-specific robot model with a parameter adaptation scheme that works well for the task but requires a lot of effort and tuning for each platform and task. In this work, we design a full model-learning-based controller based on meta pre-training that can very quickly adapt using few-shot dynamics data to any wheel-based robot with any model parameters, while also reasoning about model uncertainty. We demonstrate our results in small-scale numeric simulation, the large-scale Unity simulator, and on a medium-scale hardware platform with a wide range of settings. We show that our results are comparable to domain-specific well-engineered controllers, and have excellent generalization performance across all scenarios.
Abstract:Recent works in the robot learning community have successfully introduced generalist models capable of controlling various robot embodiments across a wide range of tasks, such as navigation and locomotion. However, achieving agile control, which pushes the limits of robotic performance, still relies on specialist models that require extensive parameter tuning. To leverage generalist-model adaptability and flexibility while achieving specialist-level agility, we propose AnyCar, a transformer-based generalist dynamics model designed for agile control of various wheeled robots. To collect training data, we unify multiple simulators and leverage different physics backends to simulate vehicles with diverse sizes, scales, and physical properties across various terrains. With robust training and real-world fine-tuning, our model enables precise adaptation to different vehicles, even in the wild and under large state estimation errors. In real-world experiments, AnyCar shows both few-shot and zero-shot generalization across a wide range of vehicles and environments, where our model, combined with a sampling-based MPC, outperforms specialist models by up to 54%. These results represent a key step toward building a foundation model for agile wheeled robot control. We will also open-source our framework to support further research.
Abstract:Off-road navigation is a challenging problem both at the planning level to get a smooth trajectory and at the control level to avoid flipping over, hitting obstacles, or getting stuck at a rough patch. There have been several recent works using classical approaches involving depth map prediction followed by smooth trajectory planning and using a controller to track it. We design an end-to-end reinforcement learning (RL) system for an autonomous vehicle in off-road environments using a custom-designed simulator in the Unity game engine. We warm-start the agent by imitating a rule-based controller and utilize Proximal Policy Optimization (PPO) to improve the policy based on a reward that incorporates Control Barrier Functions (CBF), facilitating the agent's ability to generalize effectively to real-world scenarios. The training involves agents concurrently undergoing domain-randomized trials in various environments. We also propose a novel simulation environment to replicate off-road driving scenarios and deploy our proposed approach on a real buggy RC car. Videos and additional results: https://sites.google.com/view/wroom-utd/home
Abstract:Head-to-head autonomous racing is a challenging problem, as the vehicle needs to operate at the friction or handling limits in order to achieve minimum lap times while also actively looking for strategies to overtake/stay ahead of the opponent. In this work we propose a head-to-head racing environment for reinforcement learning which accurately models vehicle dynamics. Some previous works have tried learning a policy directly in the complex vehicle dynamics environment but have failed to learn an optimal policy. In this work, we propose a curriculum learning-based framework by transitioning from a simpler vehicle model to a more complex real environment to teach the reinforcement learning agent a policy closer to the optimal policy. We also propose a control barrier function-based safe reinforcement learning algorithm to enforce the safety of the agent in a more effective way while not compromising on optimality.
Abstract:Autonomous racing is a challenging problem, as the vehicle needs to operate at the friction or handling limits in order to achieve minimum lap times. Autonomous race cars require highly accurate perception, state estimation, planning and precise application of controls. What makes it even more challenging is the accurate identification of vehicle model parameters that dictate the effects of the lateral tire slip, which may change over time, for example, due to wear and tear of the tires. Current works either propose model identification offline or need good parameters to start with (within 15-20\% of actual value), which is not enough to account for major changes in tire model that occur during actual races when driving at the control limits. We propose a unified framework which learns the tire model online from the collected data, as well as adjusts the model based on environmental changes even if the model parameters change by a higher margin. We demonstrate our approach in numeric and high-fidelity simulators for a 1:43 scale race car and a full-size car.
Abstract:Autonomous car racing is a challenging task, as it requires precise applications of control while the vehicle is operating at cornering speeds. Traditional autonomous pipelines require accurate pre-mapping, localization, and planning which make the task computationally expensive and environment-dependent. Recent works propose use of imitation and reinforcement learning to train end-to-end deep neural networks and have shown promising results for high-speed racing. However, the end-to-end models may be dangerous to be deployed on real systems, as the neural networks are treated as black-box models devoid of any provable safety guarantees. In this work we propose a decoupled approach where an optimal end-to-end controller and a state prediction end-to-end model are learned together, and the predicted state of the vehicle is used to formulate a control barrier function for safeguarding the vehicle to stay within lane boundaries. We validate our algorithm both on a high-fidelity Carla driving simulator and a 1/10-scale RC car on a real track. The evaluation results suggest that using an explicit safety controller helps to learn the task safely with fewer iterations and makes it possible to safely navigate the vehicle on the track along the more challenging racing line.
Abstract:Delays endanger safety of autonomous systems operating in a rapidly changing environment, such as nondeterministic surrounding traffic participants in autonomous driving and high-speed racing. Unfortunately, delays are typically not considered during the conventional controller design or learning-enabled controller training phases prior to deployment in the physical world. In this paper, the computation delay from nonlinear optimization for motion planning and control, as well as other unavoidable delays caused by actuators, are addressed systematically and unifiedly. To deal with all these delays, in our framework: 1) we propose a new filtering approach with no prior knowledge of dynamics and disturbance distribution to adaptively and safely estimate the time-variant computation delay; 2) we model actuation dynamics for steering delay; 3) all the constrained optimization is realized in a robust tube model predictive controller. For the application merits, we demonstrate that our approach is suitable for both autonomous driving and autonomous racing. Our approach is a novel design for a standalone delay compensation controller. In addition, in the case that a learning-enabled controller assuming no delay works as a primary controller, our approach serves as the primary controller's safety guard.
Abstract:With the rapid advancement and increased use of deep learning models in image identification, security becomes a major concern to their deployment in safety-critical systems. Since the accuracy and robustness of deep learning models are primarily attributed from the purity of the training samples, therefore the deep learning architectures are often susceptible to adversarial attacks. Adversarial attacks are often obtained by making subtle perturbations to normal images, which are mostly imperceptible to humans, but can seriously confuse the state-of-the-art machine learning models. We propose a framework, named APuDAE, leveraging Denoising AutoEncoders (DAEs) to purify these samples by using them in an adaptive way and thus improve the classification accuracy of the target classifier networks that have been attacked. We also show how using DAEs adaptively instead of using them directly, improves classification accuracy further and is more robust to the possibility of designing adaptive attacks to fool them. We demonstrate our results over MNIST, CIFAR-10, ImageNet dataset and show how our framework (APuDAE) provides comparable and in most cases better performance to the baseline methods in purifying adversaries. We also design adaptive attack specifically designed to attack our purifying model and demonstrate how our defense is robust to that.
Abstract:This is Btech thesis report on detection and purification of adverserially attacked images. A deep learning model is trained on certain training examples for various tasks such as classification, regression etc. By training, weights are adjusted such that the model performs the task well not only on training examples judged by a certain metric but has an excellent ability to generalize on other unseen examples as well which are typically called the test data. Despite the huge success of machine learning models on a wide range of tasks, security has received a lot less attention along the years. Robustness along various potential cyber attacks also should be a metric for the accuracy of the machine learning models. These cyber attacks can potentially lead to a variety of negative impacts in the real world sensitive applications for which machine learning is used such as medical and transportation systems. Hence, it is a necessity to secure the system from such attacks. Int this report, I focus on a class of these cyber attacks called the adversarial attacks in which the original input sample is modified by small perturbations such that they still look visually the same to human beings but the machine learning models are fooled by such inputs. In this report I discuss 2 novel ways to counter the adversarial attack using AutoEncoders, 1) by detecting the presence of adversaries and 2) purifying these adversaries to make target classification models robust against such attacks.