Pgx: Hardware-accelerated parallel game simulation for reinforcement learning | Sotetsu Koyamada, Shinri Okano, Soichiro Nishimori, Yu Murata, Keigo Habara, Haruka Kita, Shin Ishii | Artificial intelligence, Computer science, Deep learning, nVidia, nVidia A100, Package

Many sequential decision-making problems that are currently automated, such as those in manufacturing or recommender systems, operate in an environment where there is either little uncertainty, or zero risk of catastrophe. As companies and researchers attempt to deploy autonomous systems in less constrained environments, it is increasingly important that we endow sequential decision-making algorithms with the ability to reason about uncertainty and risk. In this thesis, we will address both planning and reinforcement learning (RL) approaches to sequential decision-making. In the planning setting, it is assumed that a model of the environment is provided, and a policy is optimised within that model. Reinforcement learning relies upon extensive random exploration, and therefore usually requires a simulator in which to perform training. In many real-world domains, it is impossible to construct a perfectly accurate model or simulator. Therefore, the performance of any policy is inevitably uncertain due to the incomplete knowledge about the environment. Furthermore, in stochastic domains, the outcome of any given run is also uncertain due to the inherent randomness of the environment. These two sources of uncertainty are usually classified as epistemic, and aleatoric uncertainty, respectively. The over-arching goal of this thesis is to contribute to developing algorithms that mitigate both sources of uncertainty in sequential decision-making problems. We make a number of contributions towards this goal, with a focus on model-based algorithms...

Recently, learning-based controllers have been shown to push mobile robotic systems to their limits and provide the robustness needed for many real-world applications. However, only classical optimization-based control frameworks offer the inherent flexibility to be dynamically adjusted during execution by, for example, setting target speeds or actuator limits. We present a framework to overcome this shortcoming of neural controllers by conditioning them on an auxiliary input. This advance is enabled by including a feature-wise linear modulation layer (FiLM). We use model-free reinforcement-learning to train quadrotor control policies for the task of navigating through a sequence of waypoints in minimum time. By conditioning the policy on the maximum available thrust or the viewing direction relative to the next waypoint, a user can regulate the aggressiveness of the quadrotor's flight during deployment. We demonstrate in simulation and in real-world experiments that a single control policy can achieve close to time-optimal flight performance across the entire performance envelope of the robot, reaching up to 60 km/h and 4.5g in acceleration. The ability to guide a learned controller during task execution has implications beyond agile quadrotor flight, as conditioning the control policy on human intent helps safely bringing learning based systems out of the well-defined laboratory environment into the wild.

This paper proposes a neural network-based user simulator that can provide a multimodal interactive environment for training Reinforcement Learning (RL) agents in collaborative tasks involving multiple modes of communication. The simulator is trained on the existing ELDERLY-AT-HOME corpus and accommodates multiple modalities such as language, pointing gestures, and haptic-ostensive actions. The paper also presents a novel multimodal data augmentation approach, which addresses the challenge of using a limited dataset due to the expensive and time-consuming nature of collecting human demonstrations. Overall, the study highlights the potential for using RL and multimodal user simulators in developing and improving domestic assistive robots.

Abstract: The local Rademacher complexity framework is one of the most successful general-purpose toolboxes for establishing sharp excess risk bounds for statistical estimators based on the framework of empirical risk minimization. Applying this toolbox typically requires using the Bernstein condition, which often restricts applicability to convex and proper settings. Recent years have witnessed several examples of problems where optimal statistical performance is only achievable via non-convex and improper estimators originating from aggregation theory, including the fundamental problem of model selection. These examples are currently outside of the reach of the classical localization theory. In this work, we build upon the recent approach to localization via offset Rademacher complexities, for which a general high-probability theory has yet to be established.

Originally published on Towards AI. I recently started an AI-focused educational newsletter, that already has over 150,000 subscribers. TheSequence is a no-BS (meaning no hype, no news etc) ML-oriented newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers and concepts. A few days ago, the data science community engaged in an intense debate when AI legend and Chief AI Scientist at Meta, Yann LeCun made some… Read the full blog for free on Medium.

Creating safe paths in unknown and uncertain environments is a challenging aspect of leader-follower formation control. In this architecture, the leader moves toward the target by taking optimal actions, and followers should also avoid obstacles while maintaining their desired formation shape. Most of the studies in this field have inspected formation control and obstacle avoidance separately. The present study proposes a new approach based on deep reinforcement learning (DRL) for end-to-end motion planning and control of under-actuated autonomous underwater vehicles (AUVs). The aim is to design optimal adaptive distributed controllers based on actor-critic structure for AUVs formation motion planning. This is accomplished by controlling the speed and heading of AUVs. In obstacle avoidance, two approaches have been deployed. In the first approach, the goal is to design control policies for the leader and followers such that each learns its own collision-free path. Moreover, the followers adhere to an overall formation maintenance policy. In the second approach, the leader solely learns the control policy, and safely leads the whole group towards the target. Here, the control policy of the followers is to maintain the predetermined distance and angle. In the presence of ocean currents, communication delays, and sensing errors, the robustness of the proposed method under realistically perturbed circumstances is shown. The efficiency of the algorithms has been evaluated and approved using a number of computer-based simulations.

Although pair trading is the simplest hedging strategy for an investor to eliminate market risk, it is still a great challenge for reinforcement learning (RL) methods to perform pair trading as human expertise. It requires RL methods to make thousands of correct actions that nevertheless have no obvious relations to the overall trading profit, and to reason over infinite states of the time-varying market most of which have never appeared in history. However, existing RL methods ignore the temporal connections between asset price movements and the risk of the performed trading. These lead to frequent tradings with high transaction costs and potential losses, which barely reach the human expertise level of trading. Therefore, we introduce CREDIT, a risk-aware agent capable of learning to exploit long-term trading opportunities in pair trading similar to a human expert. CREDIT is the first to apply bidirectional GRU along with the temporal attention mechanism to fully consider the temporal correlations embedded in the states, which allows CREDIT to capture long-term patterns of the price movements of two assets to earn higher profit. We also design the risk-aware reward inspired by the economic theory, that models both the profit and risk of the tradings during the trading period. It helps our agent to master pair trading with a robust trading preference that avoids risky trading with possible high returns and losses. Experiments show that it outperforms existing reinforcement learning methods in pair trading and achieves a significant profit over five years of U.S. stock data.

Safety is a critical component of autonomous systems and remains a challenge for learning-based policies to be utilized in the real world. In particular, policies learned using reinforcement learning often fail to generalize to novel environments due to unsafe behavior. In this paper, we propose Sim-to-Lab-to-Real to bridge the reality gap with a probabilistically guaranteed safety-aware policy distribution. To improve safety, we apply a dual policy setup where a performance policy is trained using the cumulative task reward and a backup (safety) policy is trained by solving the Safety Bellman Equation based on Hamilton-Jacobi (HJ) reachability analysis. In Sim-to-Lab transfer, we apply a supervisory control scheme to shield unsafe actions during exploration; in Lab-to-Real transfer, we leverage the Probably Approximately Correct (PAC)-Bayes framework to provide lower bounds on the expected performance and safety of policies in unseen environments. Additionally, inheriting from the HJ reachability analysis, the bound accounts for the expectation over the worst-case safety in each environment. We empirically study the proposed framework for ego-vision navigation in two types of indoor environments with varying degrees of photorealism. We also demonstrate strong generalization performance through hardware experiments in real indoor spaces with a quadrupedal robot. See https://sites.google.com/princeton.edu/sim-to-lab-to-real for supplementary material.

Highway merging scenarios featuring mixed traffic conditions pose significant modeling and control challenges for connected and automated vehicles (CAVs) interacting with incoming on-ramp human-driven vehicles (HDVs). In this paper, we present an approach to learn an approximate information state model of CAV-HDV interactions for a CAV to maneuver safely during highway merging. In our approach, the CAV learns the behavior of an incoming HDV using approximate information states before generating a control strategy to facilitate merging. First, we validate the efficacy of this framework on real-world data by using it to predict the behavior of an HDV in mixed traffic situations extracted from the Next-Generation Simulation repository. Then, we generate simulation data for HDV-CAV interactions in a highway merging scenario using a standard inverse reinforcement learning approach. Without assuming a prior knowledge of the generating model, we show that our approximate information state model learns to predict the future trajectory of the HDV using only observations. Subsequently, we generate safe control policies for a CAV while merging with HDVs, demonstrating a spectrum of driving behaviors, from aggressive to conservative. We demonstrate the effectiveness of the proposed approach by performing numerical simulations.