Providing an efficient strategy to navigate safely through unsignaled intersections is a difficult task that requires determining the intent of other drivers. We explore the effectiveness of Deep Reinforcement Learning to handle intersection problems. Using recent advances in Deep RL, we are able to learn policies that surpass the performance of a commonly-used heuristic approach in several metrics including task completion time and goal success rate and have limited ability to generalize. We then explore a system's ability to learn active sensing behaviors to enable navigating safely in the case of occlusions. Our analysis, provides insight into the intersection handling problem, the solutions learned by the network point out several shortcomings of current rule-based methods, and the failures of our current deep reinforcement learning system point to future research directions.

Deep reinforcement learning has emerged as a powerful tool for a variety of learning tasks, however deep nets typically exhibit forgetting when learning multiple tasks in sequence. To mitigate forgetting, we propose an experience replay process that augments the standard FIFO buffer and selectively stores experiences in a long-term memory. We explore four strategies for selecting which experiences will be stored: favoring surprise, favoring reward, matching the global training distribution, and maximizing coverage of the state space. We show that distribution matching successfully prevents catastrophic forgetting, and is consistently the best approach on all domains tested. While distribution matching has better and more consistent performance, we identify one case in which coverage maximization is beneficial---when tasks that receive less trained are more important. Overall, our results show that selective experience replay, when suitable selection algorithms are employed, can prevent catastrophic forgetting.

There have been numerous advances in reinforcement learning, but the typically unconstrained exploration of the learning process prevents the adoption of these methods in many safety critical applications. Recent work in safe reinforcement learning uses idealized models to achieve their guarantees, but these models do not easily accommodate the stochasticity or high-dimensionality of real world systems. We investigate how prediction provides a general and intuitive framework to constraint exploration, and show how it can be used to safely learn intersection handling behaviors on an autonomous vehicle.

Estimating statistical uncertainties allows autonomous agents to communicate their confidence during task execution and is important for applications in safety-critical domains such as autonomous driving. In this work, we present the uncertainty-aware imitation learning (UAIL) algorithm for improving end-to-end control systems via data aggregation. UAIL applies Monte Carlo Dropout to estimate uncertainty in the control output of end-to-end systems, using states where it is uncertain to selectively acquire new training data. In contrast to prior data aggregation algorithms that force human experts to visit sub-optimal states at random, UAIL can anticipate its own mistakes and switch control to the expert in order to prevent visiting a series of sub-optimal states. Our experimental results from simulated driving tasks demonstrate that our proposed uncertainty estimation method can be leveraged to reliably predict infractions. Our analysis shows that UAIL outperforms existing data aggregation algorithms on a series of benchmark tasks.

Dense urban traffic environments can produce situations where accurate prediction and dynamic models are insufficient for successful autonomous vehicle motion planning. We investigate how an autonomous agent can safely negotiate with other traffic participants, enabling the agent to handle potential deadlocks. Specifically we consider merges where the gap between cars is smaller than the size of the ego vehicle. We propose a game theoretic framework capable of generating and responding to interactive behaviors. Our main contribution is to show how game-tree decision making can be executed by an autonomous vehicle, including approximations and reasoning that make the tree-search computationally tractable. Additionally, to test our model we develop a stochastic rule-based traffic agent capable of generating interactive behaviors that can be used as a benchmark for simulating traffic participants in a crowded merge setting.

Knowledge transfer between tasks can improve the performance of learned models, but requires an accurate estimate of inter-task relationships to identify the relevant knowledge to transfer. These inter-task relationships are typically estimated based on training data for each task, which is inefficient in lifelong learning settings where the goal is to learn each consecutive task rapidly from as little data as possible. To reduce this burden, we develop a lifelong learning method based on coupled dictionary learning that utilizes high-level task descriptions to model inter-task relationships. We show that using task descriptors improves the performance of the learned task policies, providing both theoretical justification for the benefit and empirical demonstration of the improvement across a variety of learning problems. Given only the descriptor for a new task, the lifelong learner is also able to accurately predict a model for the new task through zero-shot learning using the coupled dictionary, eliminating the need to gather training data before addressing the task.

Maneuvering in dense traffic is a challenging task for autonomous vehicles because it requires reasoning about the stochastic behaviors of many other participants. In addition, the agent must achieve the maneuver within a limited time and distance. In this work, we propose a combination of reinforcement learning and game theory to learn merging behaviors. We design a training curriculum for a reinforcement learning agent using the concept of level-k behavior. This approach exposes the agent to a broad variety of behaviors during training, which promotes learning policies that are robust to model discrepancies. We show that our approach learns more efficient policies than traditional training methods.

No abstract