Simulation agents are essential for designing and testing systems that interact with humans, such as autonomous vehicles (AVs). These agents serve various purposes, from benchmarking AV performance to stress-testing system limits, but all applications share one key requirement: reliability. To enable systematic experimentation, a simulation agent must behave as intended. It should minimize actions that may lead to undesired outcomes, such as collisions, which can distort the signal-to-noise ratio in analyses. As a foundation for reliable sim agents, we propose scaling self-play to thousands of scenarios on the Waymo Open Motion Dataset under semi-realistic limits on human perception and control. Training from scratch on a single GPU, our agents nearly solve the full training set within a day. They generalize effectively to unseen test scenes, achieving a 99.8% goal completion rate with less than 0.8% combined collision and off-road incidents across 10,000 held-out scenarios. Beyond in-distribution generalization, our agents show partial robustness to out-of-distribution scenes and can be fine-tuned in minutes to reach near-perfect performance in those cases. We open-source the pre-trained agents and integrate them with a batched multi-agent simulator. Demonstrations of agent behaviors can be found at https://sites.google.com/view/reliable-sim-agents.

In the past decade, motivated by the putative failure of naive self-play deep reinforcement learning (DRL) in adversarial imperfect-information games, researchers have developed numerous DRL algorithms based on fictitious play (FP), double oracle (DO), and counterfactual regret minimization (CFR). In light of recent results of the magnetic mirror descent algorithm, we hypothesize that simpler generic policy gradient methods like PPO are competitive with or superior to these FP, DO, and CFR-based DRL approaches. To facilitate the resolution of this hypothesis, we implement and release the first broadly accessible exact exploitability computations for four large games. Using these games, we conduct the largest-ever exploitability comparison of DRL algorithms for imperfect-information games. Over 5600 training runs, FP, DO, and CFR-based approaches fail to outperform generic policy gradient methods. Code is available at https://github.com/nathanlct/IIG-RL-Benchmark and https://github.com/gabrfarina/exp-a-spiel .

Self-play has powered breakthroughs in two-player and multi-player games. Here we show that self-play is a surprisingly effective strategy in another domain. We show that robust and naturalistic driving emerges entirely from self-play in simulation at unprecedented scale -- 1.6~billion~km of driving. This is enabled by Gigaflow, a batched simulator that can synthesize and train on 42 years of subjective driving experience per hour on a single 8-GPU node. The resulting policy achieves state-of-the-art performance on three independent autonomous driving benchmarks. The policy outperforms the prior state of the art when tested on recorded real-world scenarios, amidst human drivers, without ever seeing human data during training. The policy is realistic when assessed against human references and achieves unprecedented robustness, averaging 17.5 years of continuous driving between incidents in simulation.

Designing reward functions is a core component of reinforcement learning but can be challenging for truly complex behavior. Reinforcement Learning from Human Feedback (RLHF) has been used to alleviate this challenge by replacing a hand-coded reward function with a reward function learned from preferences. However, it can be exceedingly inefficient to learn these rewards as they are often learned tabula rasa. We investigate whether Large Language Models (LLMs) can reduce this query inefficiency by converting an iterative series of human preferences into code representing the rewards. We propose In-Context Preference Learning (ICPL), a method that uses the grounding of an LLM to accelerate learning reward functions from preferences. ICPL takes the environment context and task description, synthesizes a set of reward functions, and then repeatedly updates the reward functions using human rankings of videos of the resultant policies. Using synthetic preferences, we demonstrate that ICPL is orders of magnitude more efficient than RLHF and is even competitive with methods that use ground-truth reward functions instead of preferences. Finally, we perform a series of human preference-learning trials and observe that ICPL extends beyond synthetic settings and can work effectively with humans-in-the-loop. Additional information and videos are provided at https://sites.google.com/view/few-shot-icpl/home.

A central challenge for autonomous vehicles is coordinating with humans. Therefore, incorporating realistic human agents is essential for scalable training and evaluation of autonomous driving systems in simulation. Simulation agents are typically developed by imitating large-scale, high-quality datasets of human driving. However, pure imitation learning agents empirically have high collision rates when executed in a multi-agent closed-loop setting. To build agents that are realistic and effective in closed-loop settings, we propose Human-Regularized PPO (HR-PPO), a multi-agent algorithm where agents are trained through self-play with a small penalty for deviating from a human reference policy. In contrast to prior work, our approach is RL-first and only uses 30 minutes of imperfect human demonstrations. We evaluate agents in a large set of multi-agent traffic scenes. Results show our HR-PPO agents are highly effective in achieving goals, with a success rate of 93%, an off-road rate of 3.5%, and a collision rate of 3%. At the same time, the agents drive in a human-like manner, as measured by their similarity to existing human driving logs. We also find that HR-PPO agents show considerable improvements on proxy measures for coordination with human driving, particularly in highly interactive scenarios. We open-source our code and trained agents at https://github.com/Emerge-Lab/nocturne_lab and provide demonstrations of agent behaviors at https://sites.google.com/view/driving-partners.

In this article, we explore the technical details of the reinforcement learning (RL) algorithms that were deployed in the largest field test of automated vehicles designed to smooth traffic flow in history as of 2023, uncovering the challenges and breakthroughs that come with developing RL controllers for automated vehicles. We delve into the fundamental concepts behind RL algorithms and their application in the context of self-driving cars, discussing the developmental process from simulation to deployment in detail, from designing simulators to reward function shaping. We present the results in both simulation and deployment, discussing the flow-smoothing benefits of the RL controller. From understanding the basics of Markov decision processes to exploring advanced techniques such as deep RL, our article offers a comprehensive overview and deep dive of the theoretical foundations and practical implementations driving this rapidly evolving field. We also showcase real-world case studies and alternative research projects that highlight the impact of RL controllers in revolutionizing autonomous driving. From tackling complex urban environments to dealing with unpredictable traffic scenarios, these intelligent controllers are pushing the boundaries of what automated vehicles can achieve. Furthermore, we examine the safety considerations and hardware-focused technical details surrounding deployment of RL controllers into automated vehicles. As these algorithms learn and evolve through interactions with the environment, ensuring their behavior aligns with safety standards becomes crucial. We explore the methodologies and frameworks being developed to address these challenges, emphasizing the importance of building reliable control systems for automated vehicles.

Designing traffic-smoothing cruise controllers that can be deployed onto autonomous vehicles is a key step towards improving traffic flow, reducing congestion, and enhancing fuel efficiency in mixed autonomy traffic. We bypass the common issue of having to carefully fine-tune a large traffic microsimulator by leveraging real-world trajectory data from the I-24 highway in Tennessee, replayed in a one-lane simulation. Using standard deep reinforcement learning methods, we train energy-reducing wave-smoothing policies. As an input to the agent, we observe the speed and distance of only the vehicle in front, which are local states readily available on most recent vehicles, as well as non-local observations about the downstream state of the traffic. We show that at a low 4% autonomous vehicle penetration rate, we achieve significant fuel savings of over 15% on trajectories exhibiting many stop-and-go waves. Finally, we analyze the smoothing effect of the controllers and demonstrate robustness to adding lane-changing into the simulation as well as the removal of downstream information.

A key challenge in training generally-capable agents is the design of training tasks that facilitate broad generalization and robustness to environment variations. This challenge motivates the problem setting of Unsupervised Environment Design (UED), whereby a student agent trains on an adaptive distribution of tasks proposed by a teacher agent. A pioneering approach for UED is PAIRED, which uses reinforcement learning (RL) to train a teacher policy to design tasks from scratch, making it possible to directly generate tasks that are adapted to the agent's current capabilities. Despite its strong theoretical backing, PAIRED suffers from a variety of challenges that hinder its practical performance. Thus, state-of-the-art methods currently rely on curation and mutation rather than generation of new tasks. In this work, we investigate several key shortcomings of PAIRED and propose solutions for each shortcoming. As a result, we make it possible for PAIRED to match or exceed state-of-the-art methods, producing robust agents in several established challenging procedurally-generated environments, including a partially-observed maze navigation task and a continuous-control car racing environment. We believe this work motivates a renewed emphasis on UED methods based on learned models that directly generate challenging environments, potentially unlocking more open-ended RL training and, as a result, more general agents.

We introduce Nocturne, a new 2D driving simulator for investigating multi-agent coordination under partial observability. The focus of Nocturne is to enable research into inference and theory of mind in real-world multi-agent settings without the computational overhead of computer vision and feature extraction from images. Agents in this simulator only observe an obstructed view of the scene, mimicking human visual sensing constraints. Unlike existing benchmarks that are bottlenecked by rendering human-like observations directly using a camera input, Nocturne uses efficient intersection methods to compute a vectorized set of visible features in a C++ back-end, allowing the simulator to run at over 2000 steps-per-second. Using open-source trajectory and map data, we construct a simulator to load and replay arbitrary trajectories and scenes from real-world driving data. Using this environment, we benchmark reinforcement-learning and imitation-learning agents and demonstrate that the agents are quite far from human-level coordination ability and deviate significantly from the expert trajectories.

Proximal Policy Optimization (PPO) is a popular on-policy reinforcement learning algorithm but is significantly less utilized than off-policy learning algorithms in multi-agent problems. In this work, we investigate Multi-Agent PPO (MAPPO), a multi-agent PPO variant which adopts a centralized value function. Using a 1-GPU desktop, we show that MAPPO achieves performance comparable to the state-of-the-art in three popular multi-agent testbeds: the Particle World environments, Starcraft II Micromanagement Tasks, and the Hanabi Challenge, with minimal hyperparameter tuning and without any domain-specific algorithmic modifications or architectures. In the majority of environments, we find that compared to off-policy baselines, MAPPO achieves better or comparable sample complexity as well as substantially faster running time. Finally, we present 5 factors most influential to MAPPO's practical performance with ablation studies.