Daniel Graves

Ming Zhou, Jun Luo, Julian Villella et al.

Multi-agent interaction is a fundamental aspect of autonomous driving in the real world. Despite more than a decade of research and development, the problem of how to competently interact with diverse road users in diverse scenarios remains largely unsolved. Learning methods have much to offer towards solving this problem. But they require a realistic multi-agent simulator that generates diverse and competent driving interactions. To meet this need, we develop a dedicated simulation platform called SMARTS (Scalable Multi-Agent RL Training School). SMARTS supports the training, accumulation, and use of diverse behavior models of road users. These are in turn used to create increasingly more realistic and diverse interactions that enable deeper and broader research on multi-agent interaction. In this paper, we describe the design goals of SMARTS, explain its basic architecture and its key features, and illustrate its use through concrete multi-agent experiments on interactive scenarios. We open-source the SMARTS platform and the associated benchmark tasks and evaluation metrics to encourage and empower research on multi-agent learning for autonomous driving. Our code is available at https://github.com/huawei-noah/SMARTS.

Daniel Graves, Nhat M. Nguyen, Kimia Hassanzadeh et al.

We propose a multi-time-scale predictive representation learning method to efficiently learn robust driving policies in an offline manner that generalize well to novel road geometries, and damaged and distracting lane conditions which are not covered in the offline training data. We show that our proposed representation learning method can be applied easily in an offline (batch) reinforcement learning setting demonstrating the ability to generalize well and efficiently under novel conditions compared to standard batch RL methods. Our proposed method utilizes training data collected entirely offline in the real-world which removes the need of intensive online explorations that impede applying deep reinforcement learning on real-world robot training. Various experiments were conducted in both simulator and real-world scenarios for the purpose of evaluation and analysis of our proposed claims.

Yaodong Yang, Jun Luo, Ying Wen et al.

Multiagent reinforcement learning (MARL) has achieved a remarkable amount of success in solving various types of video games. A cornerstone of this success is the auto-curriculum framework, which shapes the learning process by continually creating new challenging tasks for agents to adapt to, thereby facilitating the acquisition of new skills. In order to extend MARL methods to real-world domains outside of video games, we envision in this blue sky paper that maintaining a diversity-aware auto-curriculum is critical for successful MARL applications. Specifically, we argue that \emph{behavioural diversity} is a pivotal, yet under-explored, component for real-world multiagent learning systems, and that significant work remains in understanding how to design a diversity-aware auto-curriculum. We list four open challenges for auto-curriculum techniques, which we believe deserve more attention from this community. Towards validating our vision, we recommend modelling realistic interactive behaviours in autonomous driving as an important test bed, and recommend the SMARTS/ULTRA benchmark.

Daniel Graves, Jun Jin, Jun Luo

We propose a framework for transferring any existing policy from a potentially unknown source MDP to a target MDP. This framework (1) enables reuse in the target domain of any form of source policy, including classical controllers, heuristic policies, or deep neural network-based policies, (2) attains optimality under suitable theoretical conditions, and (3) guarantees improvement over the source policy in the target MDP. These are achieved by packaging the source policy as a black-box option in the target MDP and providing a theoretically grounded way to learn the option's initiation set through general value functions. Our approach facilitates the learning of new policies by (1) maximizing the target MDP reward with the help of the black-box option, and (2) returning the agent to states in the learned initiation set of the black-box option where it is already optimal. We show that these two variants are equivalent in performance under some conditions. Through a series of experiments in simulated environments, we demonstrate that our framework performs excellently in sparse reward problems given (sub-)optimal source policies and improves upon prior art in transfer methods such as continual learning and progressive networks, which lack our framework's desirable theoretical properties.

Jun Jin, Daniel Graves, Cameron Haigh et al.

We consider real-world reinforcement learning (RL) of robotic manipulation tasks that involve both visuomotor skills and contact-rich skills. We aim to train a policy that maps multimodal sensory observations (vision and force) to a manipulator's joint velocities under practical considerations. We propose to use offline samples to learn a set of general value functions (GVFs) that make counterfactual predictions from the visual inputs. We show that combining the offline learned counterfactual predictions with force feedbacks in online policy learning allows efficient reinforcement learning given only a terminal (success/failure) reward. We argue that the learned counterfactual predictions form a compact and informative representation that enables sample efficiency and provides auxiliary reward signals that guide online explorations towards contact-rich states. Various experiments in simulation and real-world settings were performed for evaluation. Recordings of the real-world robot training can be found via https://sites.google.com/view/realrl.

Daniel Graves, Johannes Günther, Jun Luo

General value functions (GVFs) in the reinforcement learning (RL) literature are long-term predictive summaries of the outcomes of agents following specific policies in the environment. Affordances as perceived action possibilities with specific valence may be cast into predicted policy-relative goodness and modelled as GVFs. A systematic explication of this connection shows that GVFs and especially their deep learning embodiments (1) realize affordance prediction as a form of direct perception, (2) illuminate the fundamental connection between action and perception in affordance, and (3) offer a scalable way to learn affordances using RL methods. Through an extensive review of existing literature on GVF applications and representative affordance research in robotics, we demonstrate that GVFs provide the right framework for learning affordances in real-world applications. In addition, we highlight a few new avenues of research opened up by the perspective of "affordance as GVF", including using GVFs for orchestrating complex behaviors.

Hongyao Tang, Zhaopeng Meng, Jianye Hao et al.

We study Policy-extended Value Function Approximator (PeVFA) in Reinforcement Learning (RL), which extends conventional value function approximator (VFA) to take as input not only the state (and action) but also an explicit policy representation. Such an extension enables PeVFA to preserve values of multiple policies at the same time and brings an appealing characteristic, i.e., \emph{value generalization among policies}. We formally analyze the value generalization under Generalized Policy Iteration (GPI). From theoretical and empirical lens, we show that generalized value estimates offered by PeVFA may have lower initial approximation error to true values of successive policies, which is expected to improve consecutive value approximation during GPI. Based on above clues, we introduce a new form of GPI with PeVFA which leverages the value generalization along policy improvement path. Moreover, we propose a representation learning framework for RL policy, providing several approaches to learn effective policy embeddings from policy network parameters or state-action pairs. In our experiments, we evaluate the efficacy of value generalization offered by PeVFA and policy representation learning in several OpenAI Gym continuous control tasks. For a representative instance of algorithm implementation, Proximal Policy Optimization (PPO) re-implemented under the paradigm of GPI with PeVFA achieves about 40\% performance improvement on its vanilla counterpart in most environments.

Daniel Graves, Nhat M. Nguyen, Kimia Hassanzadeh et al.

Reinforcement learning using a novel predictive representation is applied to autonomous driving to accomplish the task of driving between lane markings where substantial benefits in performance and generalization are observed on unseen test roads in both simulation and on a real Jackal robot. The novel predictive representation is learned by general value functions (GVFs) to provide out-of-policy, or counter-factual, predictions of future lane centeredness and road angle that form a compact representation of the state of the agent improving learning in both online and offline reinforcement learning to learn to drive an autonomous vehicle with methods that generalizes well to roads not in the training data. Experiments in both simulation and the real-world demonstrate that predictive representations in reinforcement learning improve learning efficiency, smoothness of control and generalization to roads that the agent was never shown during training, including damaged lane markings. It was found that learning a predictive representation that consists of several predictions over different time scales, or discount factors, improves the performance and smoothness of the control substantially. The Jackal robot was trained in a two step process where the predictive representation is learned first followed by a batch reinforcement learning algorithm (BCQ) from data collected through both automated and human-guided exploration in the environment. We conclude that out-of-policy predictive representations with GVFs offer reinforcement learning many benefits in real-world problems.

Daniel Graves, Kasra Rezaee, Sean Scheideman

We propose and demonstrate a framework called perception as prediction for autonomous driving that uses general value functions (GVFs) to learn predictions. Perception as prediction learns data-driven predictions relating to the impact of actions on the agent's perception of the world. It also provides a data-driven approach to predict the impact of the anticipated behavior of other agents on the world without explicitly learning their policy or intentions. We demonstrate perception as prediction by learning to predict an agent's front safety and rear safety with GVFs, which encapsulate anticipation of the behavior of the vehicle in front and in the rear, respectively. The safety predictions are learned through random interactions in a simulated environment containing other agents. We show that these predictions can be used to produce similar control behavior to an LQR-based controller in an adaptive cruise control problem as well as provide advanced warning when the vehicle behind is approaching dangerously. The predictions are compact policy-based predictions that support prediction of the long term impact on safety when following a given policy. We analyze two controllers that use the learned predictions in a racing simulator to understand the value of the predictions and demonstrate their use in the real-world on a Clearpath Jackal robot and an autonomous vehicle platform.

Borislav Mavrin, Daniel Graves, Alan Chan

Learning good representations is a long standing problem in reinforcement learning (RL). One of the conventional ways to achieve this goal in the supervised setting is through regularization of the parameters. Extending some of these ideas to the RL setting has not yielded similar improvements in learning. In this paper, we develop an online regularization framework for decorrelating features in RL and demonstrate its utility in several test environments. We prove that the proposed algorithm converges in the linear function approximation setting and does not change the main objective of maximizing cumulative reward. We demonstrate how to scale the approach to deep RL using the Gramian of the features achieving linear computational complexity in the number of features and squared complexity in size of the batch. We conduct an extensive empirical study of the new approach on Atari 2600 games and show a significant improvement in sample efficiency in 40 out of 49 games.

Jun Jin, Nhat M. Nguyen, Nazmus Sakib et al.

We propose a method to tackle the problem of mapless collision-avoidance navigation where humans are present using 2D laser scans. Our proposed method uses ego-safety to measure collision from the robot's perspective while social-safety to measure the impact of our robot's actions on surrounding pedestrians. Specifically, the social-safety part predicts the intrusion impact of our robot's action into the interaction area with surrounding humans. We train the policy using reinforcement learning on a simple simulator and directly evaluate the learned policy in Gazebo and real robot tests. Experiments show the learned policy can be smoothly transferred without any fine tuning. We observe that our method demonstrates time-efficient path planning behavior with high success rate in mapless navigation tasks. Furthermore, we test our method in a navigation among dynamic crowds task considering both low and high volume traffic. Our learned policy demonstrates cooperative behavior that actively drives our robot into traffic flows while showing respect to nearby pedestrians. Evaluation videos are at https://sites.google.com/view/ssw-batman

Kristopher De Asis, Alan Chan, Silviu Pitis et al.

We explore fixed-horizon temporal difference (TD) methods, reinforcement learning algorithms for a new kind of value function that predicts the sum of rewards over a $\textit{fixed}$ number of future time steps. To learn the value function for horizon $h$, these algorithms bootstrap from the value function for horizon $h-1$, or some shorter horizon. Because no value function bootstraps from itself, fixed-horizon methods are immune to the stability problems that plague other off-policy TD methods using function approximation (also known as "the deadly triad"). Although fixed-horizon methods require the storage of additional value functions, this gives the agent additional predictive power, while the added complexity can be substantially reduced via parallel updates, shared weights, and $n$-step bootstrapping. We show how to use fixed-horizon value functions to solve reinforcement learning problems competitively with methods such as Q-learning that learn conventional value functions. We also prove convergence of fixed-horizon temporal difference methods with linear and general function approximation. Taken together, our results establish fixed-horizon TD methods as a viable new way of avoiding the stability problems of the deadly triad.

Matthew Schlegel, Wesley Chung, Daniel Graves et al.

Importance sampling (IS) is a common reweighting strategy for off-policy prediction in reinforcement learning. While it is consistent and unbiased, it can result in high variance updates to the weights for the value function. In this work, we explore a resampling strategy as an alternative to reweighting. We propose Importance Resampling (IR) for off-policy prediction, which resamples experience from a replay buffer and applies standard on-policy updates. The approach avoids using importance sampling ratios in the update, instead correcting the distribution before the update. We characterize the bias and consistency of IR, particularly compared to Weighted IS (WIS). We demonstrate in several microworlds that IR has improved sample efficiency and lower variance updates, as compared to IS and several variance-reduced IS strategies, including variants of WIS and V-trace which clips IS ratios. We also provide a demonstration showing IR improves over IS for learning a value function from images in a racing car simulator.