Xue Bin Peng

Kevin Zakka, Philipp Wu, Laura Smith et al.

Replicating human-like dexterity in robot hands represents one of the largest open problems in robotics. Reinforcement learning is a promising approach that has achieved impressive progress in the last few years; however, the class of problems it has typically addressed corresponds to a rather narrow definition of dexterity as compared to human capabilities. To address this gap, we investigate piano-playing, a skill that challenges even the human limits of dexterity, as a means to test high-dimensional control, and which requires high spatial and temporal precision, and complex finger coordination and planning. We introduce RoboPianist, a system that enables simulated anthropomorphic hands to learn an extensive repertoire of 150 piano pieces where traditional model-based optimization struggles. We additionally introduce an open-sourced environment, benchmark of tasks, interpretable evaluation metrics, and open challenges for future study. Our website featuring videos, code, and datasets is available at https://kzakka.com/robopianist/

Alejandro Escontrela, Xue Bin Peng, Wenhao Yu et al.

Training a high-dimensional simulated agent with an under-specified reward function often leads the agent to learn physically infeasible strategies that are ineffective when deployed in the real world. To mitigate these unnatural behaviors, reinforcement learning practitioners often utilize complex reward functions that encourage physically plausible behaviors. However, a tedious labor-intensive tuning process is often required to create hand-designed rewards which might not easily generalize across platforms and tasks. We propose substituting complex reward functions with "style rewards" learned from a dataset of motion capture demonstrations. A learned style reward can be combined with an arbitrary task reward to train policies that perform tasks using naturalistic strategies. These natural strategies can also facilitate transfer to the real world. We build upon Adversarial Motion Priors -- an approach from the computer graphics domain that encodes a style reward from a dataset of reference motions -- to demonstrate that an adversarial approach to training policies can produce behaviors that transfer to a real quadrupedal robot without requiring complex reward functions. We also demonstrate that an effective style reward can be learned from a few seconds of motion capture data gathered from a German Shepherd and leads to energy-efficient locomotion strategies with natural gait transitions.

Yuxuan Mu, Ziyu Zhang, Yi Shi et al.

Data-driven motion priors that can guide agents toward producing naturalistic behaviors play a pivotal role in creating life-like virtual characters. Adversarial imitation learning has been a highly effective method for learning motion priors from reference motion data. However, adversarial priors, with few exceptions, need to be retrained for each new controller, thereby limiting their reusability and necessitating the retention of the reference motion data when training on downstream tasks. In this work, we present Score-Matching Motion Priors (SMP), which leverages pre-trained motion diffusion models and score distillation sampling (SDS) to create reusable task-agnostic motion priors. SMPs can be pre-trained on a motion dataset, independent of any control policy or task. Once trained, SMPs can be kept frozen and reused as general-purpose reward functions to train policies to produce naturalistic behaviors for downstream tasks. We show that a general motion prior trained on large-scale datasets can be repurposed into a variety of style-specific priors. Furthermore SMP can compose different styles to synthesize new styles not present in the original dataset. Our method produces high-quality motion comparable to state-of-the-art adversarial imitation learning methods through reusable and modular motion priors. We demonstrate the effectiveness of SMP across a diverse suite of control tasks with physically simulated humanoid characters. Video demo available at https://youtu.be/ravlZJteS20

Da Chang, Peng Xue, Yu Li et al.

Parameter-Efficient Fine-Tuning (PEFT) methods are crucial for adapting large pre-trained models. Among these, LoRA is considered a foundational approach. Building on this, the influential DoRA method enhances performance by decomposing weight updates into magnitude and direction. However, its underlying mechanism remains unclear, and it introduces significant computational overhead. In this work, we first identify that DoRA's success stems from its capacity to increase the singular value entropy of the weight update matrix, which promotes a more uniform update distribution akin to full fine-tuning. We then reformulate DoRA into a mathematically equivalent and more efficient matrix form, revealing it as a learnable weight conditioning method. Based on this insight, we propose a unified framework for designing advanced PEFT methods by exploring two orthogonal dimensions: the architectural placement and the transformation type of the conditioning matrix. Within this framework, we introduce two novel methods: (1) \textbf{Pre-Diag}, which applies a diagonal conditioning matrix before the LoRA update to efficiently calibrate the pre-trained weights, thereby enhancing performance while reducing training time; and (2) \textbf{S}kewed \textbf{O}rthogonal \textbf{R}otation \textbf{A}daptation (\textbf{SORA}), which employs a parameter-efficient orthogonal rotation to perform a more powerful, norm-preserving transformation of the feature space. Extensive experiments on natural language understanding and generation tasks demonstrate that our proposed methods achieve superior performance and efficiency compared to both LoRA and DoRA. The code is available at https://github.com/MaeChd/SORA.

Zixuan Chen, Xialin He, Yen-Jen Wang et al.

Reinforcement learning combined with sim-to-real transfer offers a general framework for developing locomotion controllers for legged robots. To facilitate successful deployment in the real world, smoothing techniques, such as low-pass filters and smoothness rewards, are often employed to develop policies with smooth behaviors. However, because these techniques are non-differentiable and usually require tedious tuning of a large set of hyperparameters, they tend to require extensive manual tuning for each robotic platform. To address this challenge and establish a general technique for enforcing smooth behaviors, we propose a simple and effective method that imposes a Lipschitz constraint on a learned policy, which we refer to as Lipschitz-Constrained Policies (LCP). We show that the Lipschitz constraint can be implemented in the form of a gradient penalty, which provides a differentiable objective that can be easily incorporated with automatic differentiation frameworks. We demonstrate that LCP effectively replaces the need for smoothing rewards or low-pass filters and can be easily integrated into training frameworks for many distinct humanoid robots. We extensively evaluate LCP in both simulation and real-world humanoid robots, producing smooth and robust locomotion controllers. All simulation and deployment code, along with complete checkpoints, is available on our project page: https://lipschitz-constrained-policy.github.io.

Seungeun Rho, Laura Smith, Tianyu Li et al.

Skill discovery methods enable agents to learn diverse emergent behaviors without explicit rewards. To make learned skills useful for unknown downstream tasks, obtaining a semantically diverse repertoire of skills is essential. While some approaches introduce a discriminator to distinguish skills and others aim to increase state coverage, no existing work directly addresses the "semantic diversity" of skills. We hypothesize that leveraging the semantic knowledge of large language models (LLMs) can lead us to improve semantic diversity of resulting behaviors. In this sense, we introduce Language Guided Skill Discovery (LGSD), a skill discovery framework that aims to directly maximize the semantic diversity between skills. LGSD takes user prompts as input and outputs a set of semantically distinctive skills. The prompts serve as a means to constrain the search space into a semantically desired subspace, and the generated LLM outputs guide the agent to visit semantically diverse states within the subspace. We demonstrate that LGSD enables legged robots to visit different user-intended areas on a plane by simply changing the prompt. Furthermore, we show that language guidance aids in discovering more diverse skills compared to five existing skill discovery methods in robot-arm manipulation environments. Lastly, LGSD provides a simple way of utilizing learned skills via natural language.

Alejandro Escontrela, Ademi Adeniji, Wilson Yan et al.

Specifying reward signals that allow agents to learn complex behaviors is a long-standing challenge in reinforcement learning. A promising approach is to extract preferences for behaviors from unlabeled videos, which are widely available on the internet. We present Video Prediction Rewards (VIPER), an algorithm that leverages pretrained video prediction models as action-free reward signals for reinforcement learning. Specifically, we first train an autoregressive transformer on expert videos and then use the video prediction likelihoods as reward signals for a reinforcement learning agent. VIPER enables expert-level control without programmatic task rewards across a wide range of DMC, Atari, and RLBench tasks. Moreover, generalization of the video prediction model allows us to derive rewards for an out-of-distribution environment where no expert data is available, enabling cross-embodiment generalization for tabletop manipulation. We see our work as starting point for scalable reward specification from unlabeled videos that will benefit from the rapid advances in generative modeling. Source code and datasets are available on the project website: https://escontrela.me/viper

Michael Laskin, Hao Liu, Xue Bin Peng et al.

We introduce Contrastive Intrinsic Control (CIC), an algorithm for unsupervised skill discovery that maximizes the mutual information between state-transitions and latent skill vectors. CIC utilizes contrastive learning between state-transitions and skills to learn behavior embeddings and maximizes the entropy of these embeddings as an intrinsic reward to encourage behavioral diversity. We evaluate our algorithm on the Unsupervised Reinforcement Learning Benchmark, which consists of a long reward-free pre-training phase followed by a short adaptation phase to downstream tasks with extrinsic rewards. CIC substantially improves over prior methods in terms of adaptation efficiency, outperforming prior unsupervised skill discovery methods by 1.79x and the next leading overall exploration algorithm by 1.18x.

Eric Mitchell, Rafael Rafailov, Xue Bin Peng et al.

This paper introduces the offline meta-reinforcement learning (offline meta-RL) problem setting and proposes an algorithm that performs well in this setting. Offline meta-RL is analogous to the widely successful supervised learning strategy of pre-training a model on a large batch of fixed, pre-collected data (possibly from various tasks) and fine-tuning the model to a new task with relatively little data. That is, in offline meta-RL, we meta-train on fixed, pre-collected data from several tasks in order to adapt to a new task with a very small amount (less than 5 trajectories) of data from the new task. By nature of being offline, algorithms for offline meta-RL can utilize the largest possible pool of training data available and eliminate potentially unsafe or costly data collection during meta-training. This setting inherits the challenges of offline RL, but it differs significantly because offline RL does not generally consider a) transfer to new tasks or b) limited data from the test task, both of which we face in offline meta-RL. Targeting the offline meta-RL setting, we propose Meta-Actor Critic with Advantage Weighting (MACAW), an optimization-based meta-learning algorithm that uses simple, supervised regression objectives for both the inner and outer loop of meta-training. On offline variants of common meta-RL benchmarks, we empirically find that this approach enables fully offline meta-reinforcement learning and achieves notable gains over prior methods.

Aviral Kumar, Xue Bin Peng, Sergey Levine

Reinforcement learning offers the promise of automating the acquisition of complex behavioral skills. However, compared to commonly used and well-understood supervised learning methods, reinforcement learning algorithms can be brittle, difficult to use and tune, and sensitive to seemingly innocuous implementation decisions. In contrast, imitation learning utilizes standard and well-understood supervised learning methods, but requires near-optimal expert data. Can we learn effective policies via supervised learning without demonstrations? The main idea that we explore in this work is that non-expert trajectories collected from sub-optimal policies can be viewed as optimal supervision, not for maximizing the reward, but for matching the reward of the given trajectory. By then conditioning the policy on the numerical value of the reward, we can obtain a policy that generalizes to larger returns. We show how such an approach can be derived as a principled method for policy search, discuss several variants, and compare the method experimentally to a variety of current reinforcement learning methods on standard benchmarks.

Xue Bin Peng, Aviral Kumar, Grace Zhang et al.

In this paper, we aim to develop a simple and scalable reinforcement learning algorithm that uses standard supervised learning methods as subroutines. Our goal is an algorithm that utilizes only simple and convergent maximum likelihood loss functions, while also being able to leverage off-policy data. Our proposed approach, which we refer to as advantage-weighted regression (AWR), consists of two standard supervised learning steps: one to regress onto target values for a value function, and another to regress onto weighted target actions for the policy. The method is simple and general, can accommodate continuous and discrete actions, and can be implemented in just a few lines of code on top of standard supervised learning methods. We provide a theoretical motivation for AWR and analyze its properties when incorporating off-policy data from experience replay. We evaluate AWR on a suite of standard OpenAI Gym benchmark tasks, and show that it achieves competitive performance compared to a number of well-established state-of-the-art RL algorithms. AWR is also able to acquire more effective policies than most off-policy algorithms when learning from purely static datasets with no additional environmental interactions. Furthermore, we demonstrate our algorithm on challenging continuous control tasks with highly complex simulated characters.

Anirudh Goyal, Shagun Sodhani, Jonathan Binas et al.

Reinforcement learning agents that operate in diverse and complex environments can benefit from the structured decomposition of their behavior. Often, this is addressed in the context of hierarchical reinforcement learning, where the aim is to decompose a policy into lower-level primitives or options, and a higher-level meta-policy that triggers the appropriate behaviors for a given situation. However, the meta-policy must still produce appropriate decisions in all states. In this work, we propose a policy design that decomposes into primitives, similarly to hierarchical reinforcement learning, but without a high-level meta-policy. Instead, each primitive can decide for themselves whether they wish to act in the current state. We use an information-theoretic mechanism for enabling this decentralized decision: each primitive chooses how much information it needs about the current state to make a decision and the primitive that requests the most information about the current state acts in the world. The primitives are regularized to use as little information as possible, which leads to natural competition and specialization. We experimentally demonstrate that this policy architecture improves over both flat and hierarchical policies in terms of generalization.

Xue Bin Peng, Angjoo Kanazawa, Sam Toyer et al.

Adversarial learning methods have been proposed for a wide range of applications, but the training of adversarial models can be notoriously unstable. Effectively balancing the performance of the generator and discriminator is critical, since a discriminator that achieves very high accuracy will produce relatively uninformative gradients. In this work, we propose a simple and general technique to constrain information flow in the discriminator by means of an information bottleneck. By enforcing a constraint on the mutual information between the observations and the discriminator's internal representation, we can effectively modulate the discriminator's accuracy and maintain useful and informative gradients. We demonstrate that our proposed variational discriminator bottleneck (VDB) leads to significant improvements across three distinct application areas for adversarial learning algorithms. Our primary evaluation studies the applicability of the VDB to imitation learning of dynamic continuous control skills, such as running. We show that our method can learn such skills directly from \emph{raw} video demonstrations, substantially outperforming prior adversarial imitation learning methods. The VDB can also be combined with adversarial inverse reinforcement learning to learn parsimonious reward functions that can be transferred and re-optimized in new settings. Finally, we demonstrate that VDB can train GANs more effectively for image generation, improving upon a number of prior stabilization methods.