Effective contact-rich manipulation requires robots to synergistically leverage vision, force, and proprioception. However, Reinforcement Learning agents struggle to learn in such multisensory settings, especially amidst sensory noise and dynamic changes. We propose MultiSensory Dynamic Pretraining (MSDP), a novel framework for learning expressive multisensory representations tailored for task-oriented policy learning. MSDP is based on masked autoencoding and trains a transformer-based encoder by reconstructing multisensory observations from only a subset of sensor embeddings, leading to cross-modal prediction and sensor fusion. For downstream policy learning, we introduce a novel asymmetric architecture, where a cross-attention mechanism allows the critic to extract dynamic, task-specific features from the frozen embeddings, while the actor receives a stable pooled representation to guide its actions. Our method demonstrates accelerated learning and robust performance under diverse perturbations, including sensor noise, and changes in object dynamics. Evaluations in multiple challenging, contact-rich robot manipulation tasks in simulation and the real world showcase the effectiveness of MSDP. Our approach exhibits strong robustness to perturbations and achieves high success rates on the real robot with as few as 6,000 online interactions, offering a simple yet powerful solution for complex multisensory robotic control.

Robots manipulating in changing environments must act on percepts that are late, noisy, or stale. We present U-LAG, a mid-execution goal-retargeting layer that leaves the low-level controller unchanged while re-aiming task goals (pre-contact, contact, post) as new observations arrive. Unlike motion retargeting or generic visual servoing, U-LAG treats in-flight goal re-aiming as a first-class, pluggable module between perception and control. Our main technical contribution is UAR-PF, an uncertainty-aware retargeter that maintains a distribution over object pose under sensing lag and selects goals that maximize expected progress. We instantiate a reproducible Shift x Lag stress test in PyBullet/PandaGym for pick, push, stacking, and peg insertion, where the object undergoes abrupt in-plane shifts while synthetic perception lag is injected during approach. Across 0-10 cm shifts and 0-400 ms lags, UAR-PF and ICP degrade gracefully relative to a no-retarget baseline, achieving higher success with modest end-effector travel and fewer aborts; simple operational safeguards further improve stability. Contributions: (1) UAR-PF for lag-adaptive, uncertainty-aware goal retargeting; (2) a pluggable retargeting interface; and (3) a reproducible Shift x Lag benchmark with evaluation on pick, push, stacking, and peg insertion.

Shared autonomy is an enabling technology that provides users with control authority over robots that would otherwise be difficult if not impossible to directly control. Yet, standard methods make assumptions that limit their adoption in practice-for example, prior knowledge of the user's goals or the objective (i.e., reward) function that they wish to optimize, knowledge of the user's policy, or query-level access to the user during training. Diffusion-based approaches to shared autonomy do not make such assumptions and instead only require access to demonstrations of desired behaviors, while allowing the user to maintain control authority. However, these advantages have come at the expense of high computational complexity, which has made real-time shared autonomy all but impossible. To overcome this limitation, we propose Consistency Shared Autonomy (CSA), a shared autonomy framework that employs a consistency model-based formulation of diffusion. Key to CSA is that it employs the distilled probability flow of ordinary differential equations (PF ODE) to generate high-fidelity samples in a single step. This results in inference speeds significantly than what is possible with previous diffusion-based approaches to shared autonomy, enabling real-time assistance in complex domains with only a single function evaluation. Further, by intervening on flawed actions at intermediate states of the PF ODE, CSA enables varying levels of assistance. We evaluate CSA on a variety of challenging simulated and real-world robot control problems, demonstrating significant improvements over state-of-the-art methods both in terms of task performance and computational efficiency.

Generalist imitation learning policies trained on large datasets show great promise for solving diverse manipulation tasks. However, to ensure generalization to different conditions, policies need to be trained with data collected across a large set of environmental factor variations (e.g., camera pose, table height, distractors) $-$ a prohibitively expensive undertaking, if done exhaustively. We introduce a principled method for deciding what data to collect and how much to collect for each factor by constructing factored scaling curves (FSC), which quantify how policy performance varies as data scales along individual or paired factors. These curves enable targeted data acquisition for the most influential factor combinations within a given budget. We evaluate the proposed method through extensive simulated and real-world experiments, across both training-from-scratch and fine-tuning settings, and show that it boosts success rates in real-world tasks in new environments by up to 26% over existing data-collection strategies. We further demonstrate how factored scaling curves can effectively guide data collection using an offline metric, without requiring real-world evaluation at scale.

Achieving zero-shot peg insertion, where inserting an arbitrary peg into an unseen hole without task-specific training, remains a fundamental challenge in robotics. This task demands a highly generalizable perception system capable of detecting potential holes, selecting the correct mating hole from multiple candidates, estimating its precise pose, and executing insertion despite uncertainties. While learning-based methods have been applied to peg insertion, they often fail to generalize beyond the specific peg-hole pairs encountered during training. Recent advancements in Vision-Language Models (VLMs) offer a promising alternative, leveraging large-scale datasets to enable robust generalization across diverse tasks. Inspired by their success, we introduce a novel zero-shot peg insertion framework that utilizes a VLM to identify mating holes and estimate their poses without prior knowledge of their geometry. Extensive experiments demonstrate that our method achieves 90.2% accuracy, significantly outperforming baselines in identifying the correct mating hole across a wide range of previously unseen peg-hole pairs, including 3D-printed objects, toy puzzles, and industrial connectors. Furthermore, we validate the effectiveness of our approach in a real-world connector insertion task on a backpanel of a PC, where our system successfully detects holes, identifies the correct mating hole, estimates its pose, and completes the insertion with a success rate of 88.3%. These results highlight the potential of VLM-driven zero-shot reasoning for enabling robust and generalizable robotic assembly.

Visual deep reinforcement learning (RL) enables robots to acquire skills from visual input for unstructured tasks. However, current algorithms suffer from low sample efficiency, limiting their practical applicability. In this work, we present MENTOR, a method that improves both the architecture and optimization of RL agents. Specifically, MENTOR replaces the standard multi-layer perceptron (MLP) with a mixture-of-experts (MoE) backbone, enhancing the agent's ability to handle complex tasks by leveraging modular expert learning to avoid gradient conflicts. Furthermore, MENTOR introduces a task-oriented perturbation mechanism, which heuristically samples perturbation candidates containing task-relevant information, leading to more targeted and effective optimization. MENTOR outperforms stateof-the-art methods across three simulation domains--DeepMind Control Suite, Meta-World, and Adroit. Additionally, MENTOR achieves an average of 83% success rate on three challenging real-world robotic manipulation tasks including Peg Insertion, Cable Routing, and Tabletop Golf, which significantly surpasses the success rate of 32% from the current strongest model-free visual RL algorithm. These results underscore the importance of sample efficiency in advancing visual RL for real-world robotics. Experimental videos are available at mentor. Figure 1: MENTOR is validated in real-world tasks. We design three challenging robotic learning tasks for the agent to acquire skills through real-world visual reinforcement learning. MENTOR achieves the most efficient and robust policies compared to the baselines. Despite substantial progress in this field (Kostrikov et al., 2020; Yarats et al., 2021; Schwarzer et al., 2020; Stooke et al., 2021; Laskin et al., 2020a), these methods still suffer from low sample efficiency.

Humans possess a remarkable talent for flexibly alternating to different senses when interacting with the environment. Picture a chef skillfully gauging the timing of ingredient additions and controlling the heat according to the colors, sounds, and aromas, seamlessly navigating through every stage of the complex cooking process. This ability is founded upon a thorough comprehension of task stages, as achieving the sub-goal within each stage can necessitate the utilization of different senses. In order to endow robots with similar ability, we incorporate the task stages divided by sub-goals into the imitation learning process to accordingly guide dynamic multi-sensory fusion. We propose MS-Bot, a stage-guided dynamic multi-sensory fusion method with coarse-to-fine stage understanding, which dynamically adjusts the priority of modalities based on the fine-grained state within the predicted current stage. We train a robot system equipped with visual, auditory, and tactile sensors to accomplish challenging robotic manipulation tasks: pouring and peg insertion with keyway. Experimental results indicate that our approach enables more effective and explainable dynamic fusion, aligning more closely with the human fusion process than existing methods.

We focus on the self-supervised discovery of manipulation concepts that can be adapted and reassembled to address various robotic tasks. We propose that the decision to conceptualize a physical procedure should not depend on how we name it (semantics) but rather on the significance of the informativeness in its representation regarding the low-level physical state and state changes. We model manipulation concepts (discrete symbols) as generative and discriminative goals and derive metrics that can autonomously link them to meaningful sub-trajectories from noisy, unlabeled demonstrations. Specifically, we employ a trainable codebook containing encodings (concepts) capable of synthesizing the end-state of a sub-trajectory given the current state (generative informativeness). Moreover, the encoding corresponding to a particular sub-trajectory should differentiate the state within and outside it and confidently predict the subsequent action based on the gradient of its discriminative score (discriminative informativeness). These metrics, which do not rely on human annotation, can be seamlessly integrated into a VQ-VAE framework, enabling the partitioning of demonstrations into semantically consistent sub-trajectories, fulfilling the purpose of discovering manipulation concepts and the corresponding sub-goal (key) states. We evaluate the effectiveness of the learned concepts by training policies that utilize them as guidance, demonstrating superior performance compared to other baselines. Additionally, our discovered manipulation concepts compare favorably to human-annotated ones while saving much manual effort.

Learning from demonstration is one of the most promising methods to counteract the challenging long-term trends in repetitive industrial assembly. It offers not only a programming technique that is accessible to workers on the shop floor, reducing the need for robot experts and the associated costs but also a possible solution to the observable shift from mass-production to mass-customisation through flexible and generalising systems. Since the emergence of the learning from demonstration idea in the 1980s, its methodologies, capabilities, and achievements have constantly evolved. However, despite reports of continued progress in academic publications, the concept has not yet robustly emerged across the assembly industry. In light of its great potential, this paper presents the findings from a systematic literature review following the updated Preferred Reporting Items for Systematic Reviews (PRISMA) guidelines. It aims to provide an overview of the state-of-the-art learning from demonstration solutions developed for assembly-related tasks and offer a critical discussion of remaining obstacles in order to drive its progression towards meaningful deployments. The analysis includes a total of 61 papers over the period of 2013-2023 sourced from Scopus and Web of Science databases. Findings indicate that learning from demonstration has attained a significant level of maturity within the research environment, as evidenced by thorough experimental achievements, proving its great promise for industrial assembly applications. However, critical obstacles exist in the area of proven practicability, task complexity and diversity, generalisation, performance evaluation and integration concepts that require attention to promote its widespread adoption and create a seamless transition into industrial practices.

A peg-in-hole insertion task is used as an example to illustrate the utility of direct associative reinforcement learning methods for learning control under real-world conditions of uncertainty and noise. Task complexity due to the use of an unchamfered hole and a clearance of less than 0.2mm is compounded by the presence of positional uncertainty of magnitude exceeding 10 to 50 times the clearance. Despite this extreme degree of uncertainty, our results indicate that direct reinforcement learning can be used to learn a robust reactive control strategy that results in skillful peg-in-hole insertions.