Roy Fox

Kolby Nottingham, Prithviraj Ammanabrolu, Alane Suhr et al. · berkeley

Reinforcement learning (RL) agents typically learn tabula rasa, without prior knowledge of the world. However, if initialized with knowledge of high-level subgoals and transitions between subgoals, RL agents could utilize this Abstract World Model (AWM) for planning and exploration. We propose using few-shot large language models (LLMs) to hypothesize an AWM, that will be verified through world experience, to improve sample efficiency of RL agents. Our DECKARD agent applies LLM-guided exploration to item crafting in Minecraft in two phases: (1) the Dream phase where the agent uses an LLM to decompose a task into a sequence of subgoals, the hypothesized AWM; and (2) the Wake phase where the agent learns a modular policy for each subgoal and verifies or corrects the hypothesized AWM. Our method of hypothesizing an AWM with LLMs and then verifying the AWM based on agent experience not only increases sample efficiency over contemporary methods by an order of magnitude but is also robust to and corrects errors in the LLM, successfully blending noisy internet-scale information from LLMs with knowledge grounded in environment dynamics.

Litian Liang, Yaosheng Xu, Stephen McAleer et al.

In temporal-difference reinforcement learning algorithms, variance in value estimation can cause instability and overestimation of the maximal target value. Many algorithms have been proposed to reduce overestimation, including several recent ensemble methods, however none have shown success in sample-efficient learning through addressing estimation variance as the root cause of overestimation. In this paper, we propose MeanQ, a simple ensemble method that estimates target values as ensemble means. Despite its simplicity, MeanQ shows remarkable sample efficiency in experiments on the Atari Learning Environment benchmark. Importantly, we find that an ensemble of size 5 sufficiently reduces estimation variance to obviate the lagging target network, eliminating it as a source of bias and further gaining sample efficiency. We justify intuitively and empirically the design choices in MeanQ, including the necessity of independent experience sampling. On a set of 26 benchmark Atari environments, MeanQ outperforms all tested baselines, including the best available baseline, SUNRISE, at 100K interaction steps in 16/26 environments, and by 68% on average. MeanQ also outperforms Rainbow DQN at 500K steps in 21/26 environments, and by 49% on average, and achieves average human-level performance using 200K ($\pm$100K) interaction steps. Our implementation is available at https://github.com/indylab/MeanQ.

Stephen McAleer, JB Lanier, Kevin Wang et al.

In competitive two-agent environments, deep reinforcement learning (RL) methods based on the \emph{Double Oracle (DO)} algorithm, such as \emph{Policy Space Response Oracles (PSRO)} and \emph{Anytime PSRO (APSRO)}, iteratively add RL best response policies to a population. Eventually, an optimal mixture of these population policies will approximate a Nash equilibrium. However, these methods might need to add all deterministic policies before converging. In this work, we introduce \emph{Self-Play PSRO (SP-PSRO)}, a method that adds an approximately optimal stochastic policy to the population in each iteration. Instead of adding only deterministic best responses to the opponent's least exploitable population mixture, SP-PSRO also learns an approximately optimal stochastic policy and adds it to the population as well. As a result, SP-PSRO empirically tends to converge much faster than APSRO and in many games converges in just a few iterations.

JB Lanier, Stephen McAleer, Pierre Baldi et al.

Robust reinforcement learning (RL) considers the problem of learning policies that perform well in the worst case among a set of possible environment parameter values. In real-world environments, choosing the set of possible values for robust RL can be a difficult task. When that set is specified too narrowly, the agent will be left vulnerable to reasonable parameter values unaccounted for. When specified too broadly, the agent will be too cautious. In this paper, we propose Feasible Adversarial Robust RL (FARR), a novel problem formulation and objective for automatically determining the set of environment parameter values over which to be robust. FARR implicitly defines the set of feasible parameter values as those on which an agent could achieve a benchmark reward given enough training resources. By formulating this problem as a two-player zero-sum game, optimizing the FARR objective jointly produces an adversarial distribution over parameter values with feasible support and a policy robust over this feasible parameter set. We demonstrate that approximate Nash equilibria for this objective can be found using a variation of the PSRO algorithm. Furthermore, we show that an optimal agent trained with FARR is more robust to feasible adversarial parameter selection than with existing minimax, domain-randomization, and regret objectives in a parameterized gridworld and three MuJoCo control environments.

Kolby Nottingham, Yasaman Razeghi, Kyungmin Kim et al.

Large language models (LLMs) are being applied as actors for sequential decision making tasks in domains such as robotics and games, utilizing their general world knowledge and planning abilities. However, previous work does little to explore what environment state information is provided to LLM actors via language. Exhaustively describing high-dimensional states can impair performance and raise inference costs for LLM actors. Previous LLM actors avoid the issue by relying on hand-engineered, task-specific protocols to determine which features to communicate about a state and which to leave out. In this work, we propose Brief Language INputs for DEcision-making Responses (BLINDER), a method for automatically selecting concise state descriptions by learning a value function for task-conditioned state descriptions. We evaluate BLINDER on the challenging video game NetHack and a robotic manipulation task. Our method improves task success rate, reduces input size and compute costs, and generalizes between LLM actors.

Kolby Nottingham, Alekhya Pyla, Sameer Singh et al.

Generalization to out of distribution tasks in reinforcement learning is a challenging problem. One successful approach improves generalization by conditioning policies on task or environment descriptions that provide information about the current transition or reward functions. Previously, these descriptions were often expressed as generated or crowd sourced text. In this work, we begin to tackle the problem of extracting useful information from natural language found in the wild (e.g. internet forums, documentation, and wikis). These natural, pre-existing sources are especially challenging, noisy, and large and present novel challenges compared to previous approaches. We propose to address these challenges by training reinforcement learning agents to learn to query these sources as a human would, and we experiment with how and when an agent should query. To address the \textit{how}, we demonstrate that pretrained QA models perform well at executing zero-shot queries in our target domain. Using information retrieved by a QA model, we train an agent to learn \textit{when} it should execute queries. We show that our method correctly learns to execute queries to maximize reward in a reinforcement learning setting.

Kyungmin Kim, JB Lanier, Pierre Baldi et al.

Recent advancements in Model-Based Reinforcement Learning (MBRL) have made it a powerful tool for visual control tasks. Despite improved data efficiency, it remains challenging to train MBRL agents with generalizable perception. Training in the presence of visual distractions is particularly difficult due to the high variation they introduce to representation learning. Building on DREAMER, a popular MBRL method, we propose a simple yet effective auxiliary task to facilitate representation learning in distracting environments. Under the assumption that task-relevant components of image observations are straightforward to identify with prior knowledge in a given task, we use a segmentation mask on image observations to only reconstruct task-relevant components. In doing so, we greatly reduce the complexity of representation learning by removing the need to encode task-irrelevant objects in the latent representation. Our method, Segmentation Dreamer (SD), can be used either with ground-truth masks easily accessible in simulation or by leveraging potentially imperfect segmentation foundation models. The latter is further improved by selectively applying the reconstruction loss to avoid providing misleading learning signals due to mask prediction errors. In modified DeepMind Control suite (DMC) and Meta-World tasks with added visual distractions, SD achieves significantly better sample efficiency and greater final performance than prior work. We find that SD is especially helpful in sparse reward tasks otherwise unsolvable by prior work, enabling the training of visually robust agents without the need for extensive reward engineering.

Armin Karamzade, Kyungmin Kim, JB Lanier et al.

Delays frequently occur in real-world environments, yet standard reinforcement learning (RL) algorithms often assume instantaneous perception of the environment. We study random sensor delays in POMDPs, where observations may arrive out-of-sequence, a setting that has not been previously addressed in RL. We analyze the structure of such delays and demonstrate that naive approaches, such as stacking past observations, are insufficient for reliable performance. To address this, we propose a model-based filtering process that sequentially updates the belief state based on an incoming stream of observations. We then introduce a simple delay-aware framework that incorporates this idea into model-based RL, enabling agents to effectively handle random delays. Applying this framework to Dreamer, we compare our approach to delay-aware baselines developed for MDPs. Our method consistently outperforms these baselines and demonstrates robustness to delay distribution shifts during deployment. Additionally, we present experiments on simulated robotic tasks, comparing our method to common practical heuristics and emphasizing the importance of explicitly modeling observation delays.

Stephen McAleer, Kevin Wang, John Lanier et al.

Policy space response oracles (PSRO) is a multi-agent reinforcement learning algorithm that has achieved state-of-the-art performance in very large two-player zero-sum games. PSRO is based on the tabular double oracle (DO) method, an algorithm that is guaranteed to converge to a Nash equilibrium, but may increase exploitability from one iteration to the next. We propose anytime double oracle (ADO), a tabular double oracle algorithm for 2-player zero-sum games that is guaranteed to converge to a Nash equilibrium while decreasing exploitability from one iteration to the next. Unlike DO, in which the restricted distribution is based on the restricted game formed by each player's strategy sets, ADO finds the restricted distribution for each player that minimizes its exploitability against any policy in the full, unrestricted game. We also propose a method of finding this restricted distribution via a no-regret algorithm updated against best responses, called RM-BR DO. Finally, we propose anytime PSRO (APSRO), a version of ADO that calculates best responses via reinforcement learning. In experiments on Leduc poker and random normal form games, we show that our methods achieve far lower exploitability than DO and PSRO and decrease exploitability monotonically.

Yaosheng Xu, Dailin Hu, Litian Liang et al.

Soft Actor-Critic (SAC) is considered the state-of-the-art algorithm in continuous action space settings. It uses the maximum entropy framework for efficiency and stability, and applies a heuristic temperature Lagrange term to tune the temperature $α$, which determines how "soft" the policy should be. It is counter-intuitive that empirical evidence shows SAC does not perform well in discrete domains. In this paper we investigate the possible explanations for this phenomenon and propose Target Entropy Scheduled SAC (TES-SAC), an annealing method for the target entropy parameter applied on SAC. Target entropy is a constant in the temperature Lagrange term and represents the target policy entropy in discrete SAC. We compare our method on Atari 2600 games with different constant target entropy SAC, and analyze on how our scheduling affects SAC.

Roy Fox, Stephen McAleer, Will Overman et al.

Multi-agent reinforcement learning has been successfully applied to fully-cooperative and fully-competitive environments, but little is currently known about mixed cooperative/competitive environments. In this paper, we focus on a particular class of multi-agent mixed cooperative/competitive stochastic games called Markov Potential Games (MPGs), which include cooperative games as a special case. Recent results have shown that independent policy gradient converges in MPGs but it was not known whether Independent Natural Policy Gradient converges in MPGs as well. We prove that Independent Natural Policy Gradient always converges in the last iterate using constant learning rates. The proof deviates from the existing approaches and the main challenge lies in the fact that Markov Potential Games do not have unique optimal values (as single-agent settings exhibit) so different initializations can lead to different limit point values. We complement our theoretical results with experiments that indicate that Natural Policy Gradient outperforms Policy Gradient in routing games and congestion games.

Ajay Kumar Tanwani, Jonathan Lee, Brijen Thananjeyan et al.

Generalizing manipulation skills to new situations requires extracting invariant patterns from demonstrations. For example, the robot needs to understand the demonstrations at a higher level while being invariant to the appearance of the objects, geometric aspects of objects such as its position, size, orientation and viewpoint of the observer in the demonstrations. In this paper, we propose an algorithm that learns a joint probability density function of the demonstrations with invariant formulations of hidden semi-Markov models to extract invariant segments (also termed as sub-goals or options), and smoothly follow the generated sequence of states with a linear quadratic tracking controller. The algorithm takes as input the demonstrations with respect to different coordinate systems describing virtual landmarks or objects of interest with a task-parameterized formulation, and adapt the segments according to the environmental changes in a systematic manner. We present variants of this algorithm in latent space with low-rank covariance decompositions, semi-tied covariances, and non-parametric online estimation of model parameters under small variance asymptotics; yielding considerably low sample and model complexity for acquiring new manipulation skills. The algorithm allows a Baxter robot to learn a pick-and-place task while avoiding a movable obstacle based on only 4 kinesthetic demonstrations.