Qiyang Li

Qiyang Li, Aviral Kumar, Ilya Kostrikov et al. · berkeley

Deep reinforcement learning algorithms that learn policies by trial-and-error must learn from limited amounts of data collected by actively interacting with the environment. While many prior works have shown that proper regularization techniques are crucial for enabling data-efficient RL, a general understanding of the bottlenecks in data-efficient RL has remained unclear. Consequently, it has been difficult to devise a universal technique that works well across all domains. In this paper, we attempt to understand the primary bottleneck in sample-efficient deep RL by examining several potential hypotheses such as non-stationarity, excessive action distribution shift, and overfitting. We perform thorough empirical analysis on state-based DeepMind control suite (DMC) tasks in a controlled and systematic way to show that high temporal-difference (TD) error on the validation set of transitions is the main culprit that severely affects the performance of deep RL algorithms, and prior methods that lead to good performance do in fact, control the validation TD error to be low. This observation gives us a robust principle for making deep RL efficient: we can hill-climb on the validation TD error by utilizing any form of regularization techniques from supervised learning. We show that a simple online model selection method that targets the validation TD error is effective across state-based DMC and Gym tasks.

Qiyang Li, Jason Zhang, Dibya Ghosh et al. · berkeley

Learning to solve tasks from a sparse reward signal is a major challenge for standard reinforcement learning (RL) algorithms. However, in the real world, agents rarely need to solve sparse reward tasks entirely from scratch. More often, we might possess prior experience to draw on that provides considerable guidance about which actions and outcomes are possible in the world, which we can use to explore more effectively for new tasks. In this work, we study how prior data without reward labels may be used to guide and accelerate exploration for an agent solving a new sparse reward task. We propose a simple approach that learns a reward model from online experience, labels the unlabeled prior data with optimistic rewards, and then uses it concurrently alongside the online data for downstream policy and critic optimization. This general formula leads to rapid exploration in several challenging sparse-reward domains where tabula rasa exploration is insufficient, including the AntMaze domain, Adroit hand manipulation domain, and a visual simulated robotic manipulation domain. Our results highlight the ease of incorporating unlabeled prior data into existing online RL algorithms, and the (perhaps surprising) effectiveness of doing so.

Qiyang Li, Yuexiang Zhai, Yi Ma et al.

Reinforcement learning (RL) problems can be challenging without well-shaped rewards. Prior work on provably efficient RL methods generally proposes to address this issue with dedicated exploration strategies. However, another way to tackle this challenge is to reformulate it as a multi-task RL problem, where the task space contains not only the challenging task of interest but also easier tasks that implicitly function as a curriculum. Such a reformulation opens up the possibility of running existing multi-task RL methods as a more efficient alternative to solving a single challenging task from scratch. In this work, we provide a theoretical framework that reformulates a single-task RL problem as a multi-task RL problem defined by a curriculum. Under mild regularity conditions on the curriculum, we show that sequentially solving each task in the multi-task RL problem is more computationally efficient than solving the original single-task problem, without any explicit exploration bonuses or other exploration strategies. We also show that our theoretical insights can be translated into an effective practical learning algorithm that can accelerate curriculum learning on simulated robotic tasks.

Qiyang Li, Ajay Jain, Pieter Abbeel

Autoregressive generative models can estimate complex continuous data distributions, like trajectory rollouts in an RL environment, image intensities, and audio. Most state-of-the-art models discretize continuous data into several bins and use categorical distributions over the bins to approximate the continuous data distribution. The advantage is that the categorical distribution can easily express multiple modes and are straightforward to optimize. However, such approximation cannot express sharp changes in density without using significantly more bins, making it parameter inefficient. We propose an efficient, expressive, multimodal parameterization called Adaptive Categorical Discretization (AdaCat). AdaCat discretizes each dimension of an autoregressive model adaptively, which allows the model to allocate density to fine intervals of interest, improving parameter efficiency. AdaCat generalizes both categoricals and quantile-based regression. AdaCat is a simple add-on to any discretization-based distribution estimator. In experiments, AdaCat improves density estimation for real-world tabular data, images, audio, and trajectories, and improves planning in model-based offline RL.

Qiyang Li, Rui Kong, Yuchen Li et al.

The integration of dynamic, sparse structures like Mixture-of-Experts (MoE) with parameter-efficient adapters (e.g., LoRA) is a powerful technique for enhancing Large Language Models (LLMs). However, this architectural enhancement comes at a steep cost: despite minimal increases in computational load, the inference latency often skyrockets, leading to decoding speeds slowing by over 2.5 times. Through a fine-grained performance analysis, we pinpoint the primary bottleneck not in the computation itself, but in the severe overhead from fragmented, sequential CUDA kernel launches required for conventional dynamic routing. To address this challenge, we introduce AdaFuse, a framework built on a tight co-design between the algorithm and the underlying hardware system to enable efficient dynamic adapter execution. Departing from conventional layer-wise or block-wise routing, AdaFuse employs a token-level pre-gating strategy, which makes a single, global routing decision for all adapter layers before a token is processed. This "decide-once, apply-everywhere" approach effectively staticizes the execution path for each token, creating an opportunity for holistic optimization. We capitalize on this by developing a custom CUDA kernel that performs a fused switching operation, merging the parameters of all selected LoRA adapters into the backbone model in a single, efficient pass. Experimental results on popular open-source LLMs show that AdaFuse achieves accuracy on par with state-of-the-art dynamic adapters while drastically cutting decoding latency by a factor of over 2.4x, thereby bridging the gap between model capability and inference efficiency.

Jin Peng Zhou, Yuhuai Wu, Qiyang Li et al.

Human mathematicians are often good at recognizing modular and reusable theorems that make complex mathematical results within reach. In this paper, we propose a novel method called theoREm-from-prooF extrACTOR (REFACTOR) for training neural networks to mimic this ability in formal mathematical theorem proving. We show on a set of unseen proofs, REFACTOR is able to extract 19.6% of the theorems that humans would use to write the proofs. When applying the model to the existing Metamath library, REFACTOR extracted 16 new theorems. With newly extracted theorems, we show that the existing proofs in the MetaMath database can be refactored. The new theorems are used very frequently after refactoring, with an average usage of 733.5 times, and help shorten the proof lengths. Lastly, we demonstrate that the prover trained on the new-theorem refactored dataset proves more test theorems and outperforms state-of-the-art baselines by frequently leveraging a diverse set of newly extracted theorems. Code can be found at https://github.com/jinpz/refactor.

Qiyang Li, Sergey Levine

We propose Q-learning with Adjoint Matching (QAM), a novel TD-based reinforcement learning (RL) algorithm that tackles a long-standing challenge in continuous-action RL: efficient optimization of an expressive diffusion or flow-matching policy with respect to a parameterized Q-function. Effective optimization requires exploiting the first-order information of the critic, but it is challenging to do so for flow or diffusion policies because direct gradient-based optimization via backpropagation through their multi-step denoising process is numerically unstable. Existing methods work around this either by only using the value and discarding the gradient information, or by relying on approximations that sacrifice policy expressivity or bias the learned policy. QAM sidesteps both of these challenges by leveraging adjoint matching, a recently proposed technique in generative modeling, which transforms the critic's action gradient to form a step-wise objective function that is free from unstable backpropagation, while providing an unbiased, expressive policy at the optimum. Combined with temporal-difference backup for critic learning, QAM consistently outperforms prior approaches on hard, sparse reward tasks in both offline and offline-to-online RL.

Max Wilcoxson, Qiyang Li, Kevin Frans et al.

Unsupervised pretraining has been transformative in many supervised domains. However, applying such ideas to reinforcement learning (RL) presents a unique challenge in that fine-tuning does not involve mimicking task-specific data, but rather exploring and locating the solution through iterative self-improvement. In this work, we study how unlabeled offline trajectory data can be leveraged to learn efficient exploration strategies. While prior data can be used to pretrain a set of low-level skills, or as additional off-policy data for online RL, it has been unclear how to combine these ideas effectively for online exploration. Our method SUPE (Skills from Unlabeled Prior data for Exploration) demonstrates that a careful combination of these ideas compounds their benefits. Our method first extracts low-level skills using a variational autoencoder (VAE), and then pseudo-labels unlabeled trajectories with optimistic rewards and high-level action labels, transforming prior data into high-level, task-relevant examples that encourage novelty-seeking behavior. Finally, SUPE uses these transformed examples as additional off-policy data for online RL to learn a high-level policy that composes pretrained low-level skills to explore efficiently. In our experiments, SUPE consistently outperforms prior strategies across a suite of 42 long-horizon, sparse-reward tasks. Code: https://github.com/rail-berkeley/supe.

Qiyang Li, Seohong Park, Sergey Levine

Temporal-difference (TD) methods learn state and action values efficiently by bootstrapping from their own future value predictions, but such a self-bootstrapping mechanism is prone to bootstrapping bias, where the errors in the value targets accumulate across steps and result in biased value estimates. Recent work has proposed to use chunked critics, which estimate the value of short action sequences ("chunks") rather than individual actions, speeding up value backup. However, extracting policies from chunked critics is challenging: policies must output the entire action chunk open-loop, which can be sub-optimal for environments that require policy reactivity and also challenging to model especially when the chunk length grows. Our key insight is to decouple the chunk length of the critic from that of the policy, allowing the policy to operate over shorter action chunks. We propose a novel algorithm that achieves this by optimizing the policy against a distilled critic for partial action chunks, constructed by optimistically backing up from the original chunked critic to approximate the maximum value achievable when a partial action chunk is extended to a complete one. This design retains the benefits of multi-step value propagation while sidestepping both the open-loop sub-optimality and the difficulty of learning action chunking policies for long action chunks. We evaluate our method on challenging, long-horizon offline goal-conditioned tasks and show that it reliably outperforms prior methods. Code: github.com/ColinQiyangLi/dqc.

Toru Lin, Yu Zhang, Qiyang Li et al.

Aiming to replicate human-like dexterity, perceptual experiences, and motion patterns, we explore learning from human demonstrations using a bimanual system with multifingered hands and visuotactile data. Two significant challenges exist: the lack of an affordable and accessible teleoperation system suitable for a dual-arm setup with multifingered hands, and the scarcity of multifingered hand hardware equipped with touch sensing. To tackle the first challenge, we develop HATO, a low-cost hands-arms teleoperation system that leverages off-the-shelf electronics, complemented with a software suite that enables efficient data collection; the comprehensive software suite also supports multimodal data processing, scalable policy learning, and smooth policy deployment. To tackle the latter challenge, we introduce a novel hardware adaptation by repurposing two prosthetic hands equipped with touch sensors for research. Using visuotactile data collected from our system, we learn skills to complete long-horizon, high-precision tasks which are difficult to achieve without multifingered dexterity and touch feedback. Furthermore, we empirically investigate the effects of dataset size, sensing modality, and visual input preprocessing on policy learning. Our results mark a promising step forward in bimanual multifingered manipulation from visuotactile data. Videos, code, and datasets can be found at https://toruowo.github.io/hato/ .

Seohong Park, Qiyang Li, Sergey Levine

We present flow Q-learning (FQL), a simple and performant offline reinforcement learning (RL) method that leverages an expressive flow-matching policy to model arbitrarily complex action distributions in data. Training a flow policy with RL is a tricky problem, due to the iterative nature of the action generation process. We address this challenge by training an expressive one-step policy with RL, rather than directly guiding an iterative flow policy to maximize values. This way, we can completely avoid unstable recursive backpropagation, eliminate costly iterative action generation at test time, yet still mostly maintain expressivity. We experimentally show that FQL leads to strong performance across 73 challenging state- and pixel-based OGBench and D4RL tasks in offline RL and offline-to-online RL. Project page: https://seohong.me/projects/fql/

Zhiyuan Zhou, Andy Peng, Qiyang Li et al.

The modern paradigm in machine learning involves pre-training on diverse data, followed by task-specific fine-tuning. In reinforcement learning (RL), this translates to learning via offline RL on a diverse historical dataset, followed by rapid online RL fine-tuning using interaction data. Most RL fine-tuning methods require continued training on offline data for stability and performance. However, this is undesirable because training on diverse offline data is slow and expensive for large datasets, and in principle, also limit the performance improvement possible because of constraints or pessimism on offline data. In this paper, we show that retaining offline data is unnecessary as long as we use a properly-designed online RL approach for fine-tuning offline RL initializations. To build this approach, we start by analyzing the role of retaining offline data in online fine-tuning. We find that continued training on offline data is mostly useful for preventing a sudden divergence in the value function at the onset of fine-tuning, caused by a distribution mismatch between the offline data and online rollouts. This divergence typically results in unlearning and forgetting the benefits of offline pre-training. Our approach, Warm-start RL (WSRL), mitigates the catastrophic forgetting of pre-trained initializations using a very simple idea. WSRL employs a warmup phase that seeds the online RL run with a very small number of rollouts from the pre-trained policy to do fast online RL. The data collected during warmup helps ``recalibrate'' the offline Q-function to the online distribution, allowing us to completely discard offline data without destabilizing the online RL fine-tuning. We show that WSRL is able to fine-tune without retaining any offline data, and is able to learn faster and attains higher performance than existing algorithms irrespective of whether they retain offline data or not.

Charles Xu, Qiyang Li, Jianlan Luo et al.

Recent advances in robotic foundation models have enabled the development of generalist policies that can adapt to diverse tasks. While these models show impressive flexibility, their performance heavily depends on the quality of their training data. In this work, we propose Reinforcement Learning Distilled Generalists (RLDG), a method that leverages reinforcement learning to generate high-quality training data for finetuning generalist policies. Through extensive real-world experiments on precise manipulation tasks like connector insertion and assembly, we demonstrate that generalist policies trained with RL-generated data consistently outperform those trained with human demonstrations, achieving up to 40% higher success rates while generalizing better to new tasks. We also provide a detailed analysis that reveals this performance gain stems from both optimized action distributions and improved state coverage. Our results suggest that combining task-specific RL with generalist policy distillation offers a promising approach for developing more capable and efficient robotic manipulation systems that maintain the flexibility of foundation models while achieving the performance of specialized controllers. Videos and code can be found on our project website https://generalist-distillation.github.io

Qiyang Li, Zhiyuan Zhou, Sergey Levine

We present Q-chunking, a simple yet effective recipe for improving reinforcement learning (RL) algorithms for long-horizon, sparse-reward tasks. Our recipe is designed for the offline-to-online RL setting, where the goal is to leverage an offline prior dataset to maximize the sample-efficiency of online learning. Effective exploration and sample-efficient learning remain central challenges in this setting, as it is not obvious how the offline data should be utilized to acquire a good exploratory policy. Our key insight is that action chunking, a technique popularized in imitation learning where sequences of future actions are predicted rather than a single action at each timestep, can be applied to temporal difference (TD)-based RL methods to mitigate the exploration challenge. Q-chunking adopts action chunking by directly running RL in a 'chunked' action space, enabling the agent to (1) leverage temporally consistent behaviors from offline data for more effective online exploration and (2) use unbiased $n$-step backups for more stable and efficient TD learning. Our experimental results demonstrate that Q-chunking exhibits strong offline performance and online sample efficiency, outperforming prior best offline-to-online methods on a range of long-horizon, sparse-reward manipulation tasks.

Perry Dong, Qiyang Li, Dorsa Sadigh et al.

We study the problem of training and fine-tuning expressive policies with online reinforcement learning (RL) given an offline dataset. Training expressive policy classes with online RL present a unique challenge of stable value maximization. Unlike simpler Gaussian policies commonly used in online RL, expressive policies like diffusion and flow-matching policies are parameterized by a long denoising chain, which hinders stable gradient propagation from actions to policy parameters when optimizing against some value function. Our key insight is that we can address stable value maximization by avoiding direct optimization over value with the expressive policy and instead construct an on-the-fly RL policy to maximize Q-value. We propose Expressive Policy Optimization (EXPO), a sample-efficient online RL algorithm that utilizes an on-the-fly policy to maximize value with two parameterized policies -- a larger expressive base policy trained with a stable imitation learning objective and a light-weight Gaussian edit policy that edits the actions sampled from the base policy toward a higher value distribution. The on-the-fly policy optimizes the actions from the base policy with the learned edit policy and chooses the value maximizing action from the base and edited actions for both sampling and temporal-difference (TD) backup. Our approach yields up to 2-3x improvement in sample efficiency on average over prior methods both in the setting of fine-tuning a pretrained policy given offline data and in leveraging offline data to train online.

Michael Janner, Qiyang Li, Sergey Levine

Reinforcement learning (RL) is typically concerned with estimating stationary policies or single-step models, leveraging the Markov property to factorize problems in time. However, we can also view RL as a generic sequence modeling problem, with the goal being to produce a sequence of actions that leads to a sequence of high rewards. Viewed in this way, it is tempting to consider whether high-capacity sequence prediction models that work well in other domains, such as natural-language processing, can also provide effective solutions to the RL problem. To this end, we explore how RL can be tackled with the tools of sequence modeling, using a Transformer architecture to model distributions over trajectories and repurposing beam search as a planning algorithm. Framing RL as sequence modeling problem simplifies a range of design decisions, allowing us to dispense with many of the components common in offline RL algorithms. We demonstrate the flexibility of this approach across long-horizon dynamics prediction, imitation learning, goal-conditioned RL, and offline RL. Further, we show that this approach can be combined with existing model-free algorithms to yield a state-of-the-art planner in sparse-reward, long-horizon tasks.

Qiyang Li, Saminul Haque, Cem Anil et al.

Lipschitz constraints under L2 norm on deep neural networks are useful for provable adversarial robustness bounds, stable training, and Wasserstein distance estimation. While heuristic approaches such as the gradient penalty have seen much practical success, it is challenging to achieve similar practical performance while provably enforcing a Lipschitz constraint. In principle, one can design Lipschitz constrained architectures using the composition property of Lipschitz functions, but Anil et al. recently identified a key obstacle to this approach: gradient norm attenuation. They showed how to circumvent this problem in the case of fully connected networks by designing each layer to be gradient norm preserving. We extend their approach to train scalable, expressive, provably Lipschitz convolutional networks. In particular, we present the Block Convolution Orthogonal Parameterization (BCOP), an expressive parameterization of orthogonal convolution operations. We show that even though the space of orthogonal convolutions is disconnected, the largest connected component of BCOP with 2n channels can represent arbitrary BCOP convolutions over n channels. Our BCOP parameterization allows us to train large convolutional networks with provable Lipschitz bounds. Empirically, we find that it is competitive with existing approaches to provable adversarial robustness and Wasserstein distance estimation.

Sicong Huang, Qiyang Li, Cem Anil et al.

In this work, we address the problem of musical timbre transfer, where the goal is to manipulate the timbre of a sound sample from one instrument to match another instrument while preserving other musical content, such as pitch, rhythm, and loudness. In principle, one could apply image-based style transfer techniques to a time-frequency representation of an audio signal, but this depends on having a representation that allows independent manipulation of timbre as well as high-quality waveform generation. We introduce TimbreTron, a method for musical timbre transfer which applies "image" domain style transfer to a time-frequency representation of the audio signal, and then produces a high-quality waveform using a conditional WaveNet synthesizer. We show that the Constant Q Transform (CQT) representation is particularly well-suited to convolutional architectures due to its approximate pitch equivariance. Based on human perceptual evaluations, we confirmed that TimbreTron recognizably transferred the timbre while otherwise preserving the musical content, for both monophonic and polyphonic samples.

Keenan Burnett, Andreas Schimpe, Sepehr Samavi et al.

The SAE AutoDrive Challenge is a three-year competition to develop a Level 4 autonomous vehicle by 2020. The first set of challenges were held in April of 2018 in Yuma, Arizona. Our team (aUToronto/Zeus) placed first. In this paper, we describe our complete system architecture and specialized algorithms that enabled us to win. We show that it is possible to develop a vehicle with basic autonomy features in just six months relying on simple, robust algorithms. We do not make use of a prior map. Instead, we have developed a multi-sensor visual localization solution. All of our algorithms run in real-time using CPUs only. We also highlight the closed-loop performance of our system in detail in several experiments.

Qiyang Li, Xintong Du, Yizhou Huang et al.

Inspired by biological swarms, robotic swarms are envisioned to solve real-world problems that are difficult for individual agents. Biological swarms can achieve collective intelligence based on local interactions and simple rules; however, designing effective distributed policies for large-scale robotic swarms to achieve a global objective can be challenging. Although it is often possible to design an optimal centralized strategy for smaller numbers of agents, those methods can fail as the number of agents increases. Motivated by the growing success of machine learning, we develop a deep learning approach that learns distributed coordination policies from centralized policies. In contrast to traditional distributed control approaches, which are usually based on human-designed policies for relatively simple tasks, this learning-based approach can be adapted to more difficult tasks. We demonstrate the efficacy of our proposed approach on two different tasks, the well-known rendezvous problem and a more difficult particle assignment problem. For the latter, no known distributed policy exists. From extensive simulations, it is shown that the performance of the learned coordination policies is comparable to the centralized policies, surpassing state-of-the-art distributed policies. Thereby, our proposed approach provides a promising alternative for real-world coordination problems that would be otherwise computationally expensive to solve or intangible to explore.

Qiyang Li, Jingxing Qian, Zining Zhu et al.

Trajectory tracking control for quadrotors is important for applications ranging from surveying and inspection, to film making. However, designing and tuning classical controllers, such as proportional-integral-derivative (PID) controllers, to achieve high tracking precision can be time-consuming and difficult, due to hidden dynamics and other non-idealities. The Deep Neural Network (DNN), with its superior capability of approximating abstract, nonlinear functions, proposes a novel approach for enhancing trajectory tracking control. This paper presents a DNN-based algorithm as an add-on module that improves the tracking performance of a classical feedback controller. Given a desired trajectory, the DNNs provide a tailored reference input to the controller based on their gained experience. The input aims to achieve a unity map between the desired and the output trajectory. The motivation for this work is an interactive "fly-as-you-draw" application, in which a user draws a trajectory on a mobile device, and a quadrotor instantly flies that trajectory with the DNN-enhanced control system. Experimental results demonstrate that the proposed approach improves the tracking precision for user-drawn trajectories after the DNNs are trained on selected periodic trajectories, suggesting the method's potential in real-world applications. Tracking errors are reduced by around 40-50% for both training and testing trajectories from users, highlighting the DNNs' capability of generalizing knowledge.