SAFE-GIL: SAFEty Guided Imitation Learning

Behavior Cloning is a popular approach to Imitation Learning, in which a robot observes an expert supervisor and learns a control policy. However, behavior cloning suffers from the "compounding error" problem - the policy errors compound as it deviates from the expert demonstrations and might lead to catastrophic system failures, limiting its use in safety-critical applications. On-policy data aggregation methods are able to address this issue at the cost of rolling out and repeated training of the imitation policy, which can be tedious and computationally prohibitive. We propose SAFE-GIL, an off-policy behavior cloning method that guides the expert via adversarial disturbance during data collection. The algorithm abstracts the imitation error as an adversarial disturbance in the system dynamics, injects it during data collection to expose the expert to safety critical states, and collects corrective actions. Our method biases training to more closely replicate expert behavior in safety-critical states and allows more variance in less critical states. We compare our method with several behavior cloning techniques and DAgger on autonomous navigation and autonomous taxiing tasks and show higher task success and safety, especially in low data regimes where the likelihood of error is higher, at a slight drop in the performance.

Imposing Exact Safety Specifications in Neural Reachable Tubes

Hamilton-Jacobi (HJ) reachability analysis is a verification tool that provides safety and performance guarantees for autonomous systems. It is widely adopted because of its ability to handle nonlinear dynamical systems with bounded adversarial disturbances and constraints on states and inputs. However, it involves solving a PDE to compute a safety value function, whose computational and memory complexity scales exponentially with the state dimension, making its direct usage in large-scale systems intractable. Recently, a learning-based approach called DeepReach, has been proposed to approximate high-dimensional reachable tubes using neural networks. While DeepReach has been shown to be effective, the accuracy of the learned solution decreases with the increase in system complexity. One of the reasons for this degradation is the inexact imposition of safety constraints during the learning process, which corresponds to the PDE's boundary conditions. Specifically, DeepReach imposes boundary conditions as soft constraints in the loss function, which leaves room for error during the value function learning. Moreover, one needs to carefully adjust the relative contributions from the imposition of boundary conditions and the imposition of the PDE in the loss function. This, in turn, induces errors in the overall learned solution. In this work, we propose a variant of DeepReach that exactly imposes safety constraints during the learning process by restructuring the overall value function as a weighted sum of the boundary condition and neural network output. This eliminates the need for a boundary loss during training, thus bypassing the need for loss adjustment. We demonstrate the efficacy of the proposed approach in significantly improving the accuracy of learned solutions for challenging high-dimensional reachability tasks, such as rocket-landing and multivehicle collision-avoidance problems.