Safety has been a key goal for autonomous driving since its inception, and we believe recognizing and responding to risk is a key component of safety. In this work, we aim to answer the question, “How can explainable risk representations be generated and used to produce risk-averse trajectories?” To answer this question, previous work uses risk metrics to formulate an optimization problem. In contrast, our work is based on research showing the usefulness of grids as a representation to generate image-based risk maps through a trained neural network. We propose a method of determining risk from a bird's eye view (BEV) of an autonomous vehicle's surroundings. Our method consists of (1) a risk map generator, which is trained to recognize risk associated with nearby agents and the map, (2) differentiable value iteration using the risk map to learn a policy, and (3) a trajectory sampler, which samples from this policy to generate a trajectory. We evaluate our planner in a close-loop manner and find improvements in its overall ability to mimic human driving while maintaining comparable safety statistics. Self-ablation also reveals the potential for fine-tuning the behavior of the planner given a designer's needs.

Many safety-critical applications of neural networks, such as robotic control, require safety guarantees. This article introduces a method for ensuring the safety of learned models for control using differentiable control barrier functions (dCBFs). dCBFs are end-to-end trainable and guarantee safety. They improve over classical control barrier functions (CBFs), which are usually overly conservative. Our dCBF solution relaxes the CBF definitions by: 1) using environmental dependencies; 2) embedding them into differentiable quadratic programs. These novel safety layers are called a BarrierNet. They can be used in conjunction with any neural network-based controller. They are trained by gradient descent. With BarrierNet, the safety constraints of a neural controller become adaptable to changing environments. We evaluate BarrierNet on the following several problems: 1) robot traffic merging; 2) robot navigation in 2-D and 3-D spaces; 3) end-to-end vision-based autonomous driving in a sim-to-real environment and in physical experiments; 4) demonstrate their effectiveness compared to state-of-the-art approaches.

Temporal patterns (how vehicles behave in our observed past) underline our reasoning of how people drive on the road, and can explain why we make certain predictions about interactions among road agents. In this paper we propose the ConceptNet trajectory predictor - a novel prediction framework that is able to incorporate agent interactions as explicit edges in a temporal knowledge graph. We demonstrate the sample efficiency and the overall accuracy of the proposed approach, and show that using the graphical structure to explicitly model interactions enables better detection of agent interactions and improved trajectory predictions on a large real-world driving dataset.

Learning composable policies for environments with complex rules and tasks is a challenging problem. We introduce a hierarchical reinforcement learning framework called the Logical Options Framework (LOF) that learns policies that are satisfying, optimal, and composable. LOF efficiently learns policies that satisfy tasks by representing the task as an automaton and integrating it into learning and planning. We provide and prove conditions under which LOF will learn satisfying, optimal policies. And lastly, we show how LOF’s learned policies can be composed to satisfy unseen tasks with only 10-50 retraining steps on our benchmarks. We evaluate LOF on four tasks in discrete and continuous domains, including a 3D pick-and-place environment.