Language-conditioned robotic manipulation in unstructured environments presents significant challenges for intelligent robotic systems. However, due to partial observation or imprecise action prediction, failure may be unavoidable for learned policies. Moreover, operational failures can lead to the robotic arm entering an untrained state, potentially causing destructive results. Consequently, the ability to detect and self-correct failures is crucial for the development of practical robotic systems. To address this challenge, we propose a foresight-driven failure detection and self-correction module for robot manipulation. By leveraging 3D Gaussian Splatting, we represent the current scene with multiple Gaussians. Subsequently, we train a prediction network to forecast the Gaussian representation of future scenes conditioned on planned actions. Failure is detected when the predicted future significantly deviates from the real observation after action execution. In such cases, the end-effector rolls back to the previous action to avoid an untrained state. Integrating this approach with the PerACT framework, we develop a self-correcting robot manipulation policy. Evaluations on ten RLBench tasks with 166 variations demonstrate the superior performance of the proposed method, which outperforms state-of-the-art methods by 12.0% success rate on average.
论文检索
输入标题、作者或关键词，从 2,942 篇学术成果中精准定位
Assignment problems are a classic combinatorial optimization problem in which a group of agents must be assigned to a group of tasks such that maximum utility is achieved while satisfying assignment constraints. Given the utility of each agent completing each task, polynomial-time algorithms exist to solve a single assignment problem in its simplest form. However, in many modern-day applications such as satellite constellations, power grids, and mobile robot scheduling, assignment problems unfold over time, with the utility for a given assignment depending heavily on the state of the system. We apply multi-agent reinforcement learning to this problem, learning the value of assignments by bootstrapping from the known polynomial-time greedy solver and then learning from further experience. We then choose assignments using a distributed optimal assignment mechanism rather than by selecting them directly. We demonstrate that this algorithm is theoretically justified and avoids pitfalls experienced by other RL algorithms in this setting. Finally, we show that our algorithm significantly outperforms other methods in the literature, even while scaling to realistic scenarios with hundreds of agents and tasks.
Multi-Agent Motion Planning (MAMP) finds various applications in fields such as traffic management, airport operations, and warehouse automation. In many of these environments, differential drive robots are commonly used. These robots have a kinodynamic model that allows only in-place rotation and movement along their current orientation, subject to speed and acceleration limits. However, existing Multi-Agent Path Finding (MAPF)-based methods often use simplified models for robot kinodynamics, which limits their practicality and realism. In this paper, we introduce a three-level framework called MASS to address these challenges. MASS combines MAPF-based methods with our proposed stationary state search planner to generate high-quality kinodynamically-feasible plans. We further extend MASS using an adaptive window mechanism to address the lifelong MAMP problem. Empirically, we tested our methods on the single-shot grid map domain and the lifelong warehouse domain. Our method shows up to 400% improvements in terms of throughput compared to existing methods.
LNS2+RL: Combining Multi-agent Reinforcement Learning with Large Neighborhood Search in Multi-agent Path Finding
PDF ↗Multi-Agent Path Finding (MAPF) is a critical component of logistics and warehouse management, which focuses on planning collision-free paths for a team of robots in a known environment. Recent work introduced a novel MAPF approach, LNS2, which proposed to repair a quickly-obtainable set of infeasible paths via iterative re-planning, by relying on a fast, yet lower-quality, priority-based planner. At the same time, there has been a recent push for Multi-Agent Reinforcement Learning (MARL) based MAPF algorithms, which let agents learn decentralized policies that exhibit improved cooperation over such priority planning, although inevitably remaining slower. In this paper, we introduce a new MAPF algorithm, LNS2+RL, which combines the distinct yet complementary characteristics of LNS2 and MARL to effectively balance their individual limitations and get the best from both worlds. During early iterations, LNS2+RL relies on MARL for low-level re-planning, which we show eliminates collisions much more than a priority-based planner. There, our MARL-based planner allows agents to reason about past and future/predicted information to gradually learn cooperative decision-making through a finely designed curriculum learning. At later stages of planning, LNS2+RL adaptively switches to priority-based planning to quickly resolve the remaining collisions, naturally trading-off solution quality and computational efficiency. Our comprehensive experiments on challenging tasks across various team sizes, world sizes, and map structures consistently demonstrate the superior performance of LNS2+RL compared to many MAPF algorithms, including LNS2, LaCAM, and EECBS. In maps with complex structures, the advantages of LNS2+RL are particularly pronounced, with LNS2+RL achieving a success rate of over 50% in nearly half of the tested tasks, while that of LaCAM and EECBS falls to 0%.
Differentiable environments have heralded new possibilities for learning control policies by offering rich differentiable information that facilitates gradient-based methods. In comparison to prevailing model-free reinforcement learning approaches, model-based reinforcement learning (MBRL) methods exhibit the potential to effectively harness the power of differentiable information for recovering the underlying physical dynamics. However, this presents two primary challenges: effectively utilizing differentiable information to 1) construct models with more accurate dynamic prediction and 2) enhance the stability of policy training. In this paper, we propose a Differentiable Information Enhanced MBRL method, MB-MIX, to address both challenges. Firstly, we adopt a Sobolev model training approach that penalizes incorrect model gradient outputs, enhancing prediction accuracy and yielding more precise models that faithfully capture system dynamics. Secondly, we introduce mixing lengths of truncated learning windows to reduce the variance in policy gradient estimation, resulting in improved stability during policy learning. To validate the effectiveness of our approach in differentiable environments, we provide theoretical analysis and empirical results. Notably, our approach outperforms previous model-based and model-free methods, in multiple challenging tasks involving controllable rigid robots such as humanoid robots' motion control and deformable object manipulation.
In this work, we present an in-context policy adaptation (ICPAD) framework designed for long-horizon multi-task environments, exploring diffusion-based skill learning techniques in cross-domain settings. The framework enables rapid adaptation of skill-based reinforcement learning policies to diverse target domains, especially under stringent constraints on no model updates and only limited target domain data. Specifically, the framework employs a cross-domain skill diffusion scheme, where domain-agnostic prototype skills and a domain-grounded skill adapter are learned jointly and effectively from an offline dataset through cross-domain consistent diffusion processes. The prototype skills act as primitives for common behavior representations of long-horizon policies, serving as a lingua franca to bridge different domains. Furthermore, to enhance the in-context adaptation performance, we develop a dynamic domain prompting scheme that guides the diffusion-based skill adapter toward better alignment with the target domain. Through experiments with robotic manipulation in Metaworld and autonomous driving in CARLA, we show that our ICPAD framework achieves superior policy adaptation performance under limited target domain data conditions for various cross-domain configurations including differences in environment dynamics, agent embodiment, and task horizon.
Deep reinforcement learning (DRL) has achieved remarkable success in various domains, yet its reliance on neural networks results in a lack of transparency, which limits its practical applications in safety-critical and human-agent interaction domains. Decision trees, known for their notable explainability, have emerged as a promising alternative to neural networks. However, decision trees often struggle in long-horizon continuous control tasks with high-dimensional observation space due to their limited expressiveness. To address this challenge, we propose SkillTree, a novel hierarchical framework that reduces the complex continuous action space of challenging control tasks into discrete skill space. By integrating the differentiable decision tree within the high-level policy, SkillTree generates discrete skill embeddings that guide low-level policy execution. Furthermore, through distillation, we obtain a simplified decision tree model that improves performance while further reducing complexity. Experiment results validate SkillTree’s effectiveness across various robotic manipulation tasks, providing clear skill-level insights into the decision-making process. The proposed approach not only achieves performance comparable to neural network based methods in complex long-horizon control tasks but also significantly enhances the transparency and explainability of the decision-making process.
We study the problem of computing deterministic optimal policies for constrained Markov decision processes (MDPs) with continuous state and action spaces, which are widely encountered in constrained dynamical systems. Designing deterministic policy gradient methods in continuous state and action spaces is particularly challenging due to the lack of enumerable state-action pairs and the adoption of deterministic policies, hindering the application of existing policy gradient methods for constrained MDPs. To this end, we develop a deterministic policy gradient primal-dual method to find an optimal deterministic policy with non-asymptotic convergence. Specifically, we leverage regularization of the Lagrangian of the constrained MDP to propose a deterministic policy gradient primal-dual (D-PGPD) algorithm that updates the deterministic policy via a quadratic-regularized gradient ascent step and the dual variable via a quadratic-regularized gradient descent step. We prove that the primal-dual iterates of D-PGPD converge at a sub-linear rate to an optimal regularized primal-dual pair. We instantiate D-PGPD with function approximation and prove that the primal-dual iterates of D-PGPD converge at a sub-linear rate to an optimal regularized primal-dual pair, up to a function approximation error. Furthermore, we demonstrate the effectiveness of our method in two continuous control problems: robot navigation and fluid control. To the best of our knowledge, this appears to be the first work that proposes a deterministic policy search method for continuous-space constrained MDPs.
This study challenges strictly guaranteeing ``dissipativity'' of a dynamical system represented by neural networks learned from given time-series data. Dissipativity is a crucial indicator for dynamical systems that generalizes stability and input-output stability, known to be valid across various systems including robotics, biological systems, and molecular dynamics. By analytically proving the general solution to the nonlinear Kalman–Yakubovich–Popov (KYP) lemma, which is the necessary and sufficient condition for dissipativity, we propose a differentiable projection that transforms any dynamics represented by neural networks into dissipative ones and a learning method for the transformed dynamics. Utilizing the generality of dissipativity, our method strictly guarantee stability, input-output stability, and energy conservation of trained dynamical systems. Finally, we demonstrate the robustness of our method against out-of-domain input through applications to robotic arms and fluid dynamics.
Multi-objective decision-making problems have emerged in numerous real-world scenarios, such as video games, navigation and robotics. Considering the clear advantages of Reinforcement Learning (RL) in optimizing decision-making processes, researchers have delved into the development of Multi-Objective RL (MORL) methods for solving multi-objective decision problems. However, previous methods either cannot obtain the entire Pareto front, or employ only a single policy network for all the preferences over multiple objectives, which may not produce personalized solutions for each preference. To address these limitations, we propose a novel decomposition-based framework for MORL, Pareto Set Learning for MORL (PSL-MORL), that harnesses the generation capability of hypernetwork to produce the parameters of the policy network for each decomposition weight, generating relatively distinct policies for various scalarized subproblems with high efficiency. PSL-MORL is a general framework, which is compatible for any RL algorithm. The theoretical result guarantees the superiority of the model capacity of PSL-MORL and the optimality of the obtained policy network. Through extensive experiments on diverse benchmarks, we demonstrate the effectiveness of PSL-MORL in achieving dense coverage of the Pareto front, significantly outperforming state-of-the-art MORL methods in both the hypervolume and sparsity indicators.
In this paper, we study reinforcement learning in Markov Decision Processes with Probabilistic Reward Machines (PRMs), a form of non-Markovian reward commonly found in robotics tasks. We design an algorithm for PRMs that achieves a regret bound of Õ((HOAT)^(1/2) + H²O²A^(3/2) + H(T)^(1/2)), where H is the time horizon, O is the number of observations, A is the number of actions, and T is the number of time steps. This result improves over the best-known bound, Õ(H(OAT)^(1/2)), for MDPs with Deterministic Reward Machines (DRMs), a special case of PRMs. When T ≥ H³O³A² and OA ≥ H, our regret bound leads to a regret of Õ((HOAT)^(1/2)), which matches the established lower bound of Ω((HOAT)^(1/2)) for MDPs with DRMs up to a logarithmic factor. To the best of our knowledge, this is the first efficient algorithm for PRMs. Additionally, we present a new simulation lemma for non-Markovian rewards, which enables reward-free exploration for any non-Markovian reward given access to an approximate planner. Complementing our theoretical findings, we show through extensive experimental evaluations that our algorithm indeed outperforms prior methods in various PRM environments.
Representation learning is a powerful tool that enables learning over large multitudes of agents or domains by enforcing that all agents operate on a shared set of learned features. However, many robotics or controls applications that would benefit from collaboration operate in settings with changing environments and goals, whereas most guarantees for representation learning are stated for static settings. Toward rigorously establishing the benefit of representation learning in dynamic settings, we analyze the regret of multi-task representation learning for linear-quadratic control. This setting introduces unique challenges. Firstly, we must account for and balance the misspecification introduced by an approximate representation. Secondly, we cannot rely on the parameter update schemes of single-task online LQR, for which least-squares often suffices, and must devise a novel scheme to ensure sufficient improvement. We demonstrate that for settings where exploration is "benign", the regret of any agent after T timesteps scales with the square root of T/H, where H is the number of agents. In settings with "difficult" exploration, the regret scales as the square root of the input dimension times the parameter dimension multiplied by T, plus a term which scales with T to the three quarters divided by H to the one fifth. In both cases, by comparing to the minimax single-task regret, we see a benefit of a large number of agents. Notably, in the difficult exploration case, by sharing a representation across tasks, the effective task-specific parameter count can often be small. Lastly, we validate the trends we predict.
Bayesian optimization (BO) is a key technique for solving black-box optimization problems. This study extends the scope of BO from conventional applications (e.g., AutoML and robotics learning) to automated tuning of software systems. Despite GP (Gaussian Process) implementing a foundation formalism for exploitation and exploration in BO, its limited predictive power and unrealistic assumptions (e.g., continuity and Gaussianity) can severely affect its effectiveness and efficiency in tuning complex software systems. To overcome these limitations, we propose a BO framework CoffeeBoost, which implements exploitation and exploration with a GBDT-native distribution-free probabilistic surrogate model. CoffeeBoost constructs surrogate models via stochastic gradient boosting ensembles (SGBE) and quantifies probabilistic distributions via distribution-free conformal predictive systems. Moreover, CoffeeBoost leverages the residual paths in SGBE to improve the local adaptiveness of the resulting predictive distributions in a GBDT-native manner. Across eight auto-tuning benchmarks for database management systems (DBMS), we evaluate CoffeeBoost and show its superior learnability and optimizability against existing GP-based and tree-ensemble-based BO schemes. Detailed analysis further shows CoffeeBoost's predictive distributions excel in both coverage and tightness.
Importance sampling is a rare event simulation technique used in Monte Carlo simulations to bias the sampling distribution towards the rare event of interest. By assigning appropriate weights to sampled points, importance sampling allows for more efficient estimation of rare events or tails of distributions. However, importance sampling can fail when the proposal distribution does not effectively cover the target distribution. In this work, we propose a method for more efficient sampling by updating the proposal distribution in the latent space of a normalizing flow. Normalizing flows learn an invertible mapping from a target distribution to a simpler latent distribution. The latent space can be more easily explored during the search for a proposal distribution, and samples from the proposal distribution are recovered in the space of the target distribution via the invertible mapping. We empirically validate our methodology on simulated robotics applications such as autonomous racing and aircraft ground collision avoidance.
M3Net: Efficient Time-Frequency Integration Network with Mirror Attention for Audio Classification on Edge
PDF ↗Audio classification plays a crucial role within fields such as human-machine interaction and intelligent robotics. However, high-performance audio classification systems typically demand significant computational and storage resources, posing substantial challenges when deploying to the resource-constrained edge devices with an urgent need for such capabilities. To achieve a new level of balance between model complexity and performance, we introduce a novel multi-view method for the separated time-frequency features extraction and utilization, which exists within the proposed Mini Mirror Multi-View Network (M3Net) in the form of the Mirror Attention mechanism. M3Net enables reversible spatial transformation of spectral features is capable of efficiently leverages robust local and global features in the time and frequency domains with low requirements for parameters. Experiments based on Mel-Spectrogram without data augmentation and pre-training indicate that M3Net can achieve classification accuracy over 97% on the UrbanSound8K and SpeechCommandsV2 datasets with only 0.03 million parameters. The contribution of each functional segment in M3Net is fully verified and explained in the ablation experiments.
The scalability of instructable agents in robotics or gaming is often hindered by limited data that pairs instructions with agent trajectories. However, large datasets of unannotated trajectories containing sequences of various agent behaviour (play trajectories) are often available. In a semi-supervised setup, we explore methods to extract labelled segments from play trajectories. The goal is to augment a small annotated dataset of instruction-trajectory pairs to improve the performance of an instruction-following policy trained downstream via imitation learning. Assuming little variation in segment length, recent video segmentation methods can effectively extract labelled segments. To address the constraint of segment length, we propose Play Segmentation (PS), a probabilistic model that finds maximum likely segmentations of extended subsegments, while only being trained on individual instruction segments. Our results in a game environment and a simulated robotic gripper setting underscore the importance of segmentation; randomly sampled segments diminish performance, while incorporating labelled segments from PS improves policy performance to the level of a policy trained on twice the amount of labelled data.
This paper presents a novel framework for multi-armed bandit problems with side-observations and switching constraints, which arises in a range of real-world applications such as robotic. To address the challenges of effectively utilizing graph-structured observations while adhering to graph constraints, we design graph-agnostic and graph-aware algorithms tailored to this new setting. Specifically, our graph-agnostic algorithm selects nodes with the highest upper confidence bound without prior knowledge of feedback probabilities, while minimizing switching costs using offline shortest path planning and the doubling trick. If the graph structure and associated probability matrix are known, our graph-aware algorithm plans the exploration step using a linear programming approach and eliminates suboptimal nodes iteratively. We rigorously analyze the performance of our proposed algorithms, providing near-optimal minimax and instance-dependent regret upper bounds. Our analysis shows that our algorithms outperform generic reinforcement learning methods in terms of both regret and computational efficiency. Extensive numerical experiments on various types of graphs, including two real-world datasets, demonstrate the efficacy of our proposed methods and their advantages over benchmark methods in graph bandit settings.
DUO: Diverse, Uncertain, On-Policy Query Generation and Selection for Reinforcement Learning from Human Feedback
PDF ↗Defining a reward function is usually a challenging but critical task for the system designer in reinforcement learning, especially when specifying complex behaviors. Reinforcement learning from human feedback (RLHF) emerges as a promising approach to circumvent this. In RLHF, the agent typically learns a reward function by querying a human teacher using pairwise comparisons of trajectory segments. A key question in this domain is how to reduce the number of queries necessary to learn an informative reward function since asking a human teacher too many queries is impractical and costly. To tackle this question, we propose DUO, a novel method for diverse, uncertain, on-policy query generation and selection in RLHF. Our method produces queries that are (1) more relevant for policy training (via an on-policy criterion), (2) more informative (via a principled measure of epistemic uncertainty), and (3) diverse (via a clustering-based filter). Experimental results on a variety of locomotion and robotic manipulation tasks demonstrate that our method can outperform state-of-the-art RLHF methods given the same total budget of queries, while being robust to possibly irrational teachers.
Evaluating deep reinforcement learning (DRL) agents against targeted behavior attacks is critical for assessing their robustness. These attacks aim to manipulate the victim into specific behaviors that align with the attacker’s objectives, often bypassing traditional reward-based defenses. Prior methods have primarily focused on reducing cumulative rewards; however, rewards are typically too generic to capture complex safety requirements effectively. As a result, focusing solely on reward reduction can lead to suboptimal attack strategies, particularly in safety-critical scenarios where more precise behavior manipulation is needed. To address these challenges, we propose RAT, a method designed for universal, targeted behavior attacks. RAT trains an intention policy that is explicitly aligned with human preferences, serving as a precise behavioral target for the adversary. Concurrently, an adversary manipulates the victim's policy to follow this target behavior. To enhance the effectiveness of these attacks, RAT dynamically adjusts the state occupancy measure within the replay buffer, allowing for more controlled and effective behavior manipulation. Our empirical results on robotic simulation tasks demonstrate that RAT outperforms existing adversarial attack algorithms in inducing specific behaviors. Additionally, RAT shows promise in improving agent robustness, leading to more resilient policies. We further validate RAT by guiding Decision Transformer agents to adopt behaviors aligned with human preferences in various MuJoCo tasks, demonstrating its effectiveness across diverse tasks.
Imitation learning (IL) is notably effective for robotic tasks where directly programming behaviors or defining optimal control costs is challenging. In this work, we address a scenario where the imitator relies solely on observed behavior and cannot make environmental interactions during learning. It does not have additional supplementary datasets beyond the expert's dataset nor any information about the transition dynamics. Unlike state-of-the-art (SOTA) IL methods, this approach tackles the limitations of conventional IL by operating in a more constrained and realistic setting. Our method uses the Markov balance equation and introduces a novel conditional density estimation-based imitation learning framework. It employs conditional normalizing flows for transition dynamics estimation and aims at satisfying a balance equation for the environment. Through a series of numerical experiments on Classic Control and MuJoCo environments, we demonstrate consistently superior empirical performance compared to many SOTA IL algorithms.