RFS: Reinforcement learning with Residual flow steering for dexterous manipulation

Imitation learning has emerged as an effective approach for bootstrapping sequential decision-making in robotics, achieving strong performance even in high-dimensional dexterous manipulation tasks. Recent behavior cloning methods further leverage expressive generative models, such as diffusion models and flow matching, to represent multimodal action distributions. However, policies pretrained in this manner often exhibit limited generalization and require additional fine-tuning to achieve robust performance at deployment time. Such adaptation must preserve the global exploration benefits of pretraining while enabling rapid correction of local execution errors. We propose Residual Flow Steering(RFS), a data-efficient reinforcement learning framework for adapting pretrained generative policies. RFS steers a pretrained flow-matching policy by jointly optimizing a residual action and a latent noise distribution, enabling complementary forms of exploration: local refinement through residual corrections and global exploration through latent-space modulation. This design allows efficient adaptation while retaining the expressive structure of the pretrained policy. We demonstrate the effectiveness of RFS on dexterous manipulation tasks, showing efficient fine-tuning in both simulation and real-world settings when adapting pretrained base policies. Project website:https://weirdlabuw.github.io/rfs.

Hierarchical Proportion Models for Motion Generation via Integration of Motion Primitives

Imitation learning (IL) enables robots to acquire human-like motion skills from demonstrations, but it still requires extensive high-quality data and retraining to handle complex or long-horizon tasks. To improve data efficiency and adaptability, this study proposes a hierarchical IL framework that integrates motion primitives with proportion-based motion synthesis. The proposed method employs a two-layer architecture, where the upper layer performs long-term planning, while a set of lower-layer models learn individual motion primitives, which are combined according to specific proportions. Three model variants are introduced to explore different trade-offs between learning flexibility, computational cost, and adaptability: a learning-based proportion model, a sampling-based proportion model, and a playback-based proportion model, which differ in how the proportions are determined and whether the upper layer is trainable. Through real-robot pick-and-place experiments, the proposed models successfully generated complex motions not included in the primitive set. The sampling-based and playback-based proportion models achieved more stable and adaptable motion generation than the standard hierarchical model, demonstrating the effectiveness of proportion-based motion integration for practical robot learning.

On the Sample Efficiency of Inverse Dynamics Models for Semi-Supervised Imitation Learning

Semi-supervised imitation learning (SSIL) consists in learning a policy from a small dataset of action-labeled trajectories and a much larger dataset of action-free trajectories. Some SSIL methods learn an inverse dynamics model (IDM) to predict the action from the current state and the next state. An IDM can act as a policy when paired with a video model (VM-IDM) or as a label generator to perform behavior cloning on action-free data (IDM labeling). In this work, we first show that VM-IDM and IDM labeling learn the same policy in a limit case, which we call the IDM-based policy. We then argue that the previously observed advantage of IDM-based policies over behavior cloning is due to the superior sample efficiency of IDM learning, which we attribute to two causes: (i) the ground-truth IDM tends to be contained in a lower complexity hypothesis class relative to the expert policy, and (ii) the ground-truth IDM is often less stochastic than the expert policy. We argue these claims based on insights from statistical learning theory and novel experiments, including a study of IDM-based policies using recent architectures for unified video-action prediction (UVA). Motivated by these insights, we finally propose an improved version of the existing LAPO algorithm for latent action policy learning.

PRISM: Performer RS-IMLE for Single-pass Multisensory Imitation Learning

Robotic imitation learning typically requires models that capture multimodal action distributions while operating at real-time control rates and accommodating multiple sensing modalities. Although recent generative approaches such as diffusion models, flow matching, and Implicit Maximum Likelihood Estimation (IMLE) have achieved promising results, they often satisfy only a subset of these requirements. To address this, we introduce PRISM, a single-pass policy based on a batch-global rejection-sampling variant of IMLE. PRISM couples a temporal multisensory encoder (integrating RGB, depth, tactile, audio, and proprioception) with a linear-attention generator using a Performer architecture. We demonstrate the efficacy of PRISM on a diverse real-world hardware suite, including loco-manipulation using a Unitree Go2 with a 7-DoF arm D1 and tabletop manipulation with a UR5 manipulator. Across challenging physical tasks such as pre-manipulation parking, high-precision insertion, and multi-object pick-and-place, PRISM outperforms state-of-the-art diffusion policies by 10-25% in success rate while maintaining high-frequency (30-50 Hz) closed-loop control. We further validate our approach on large-scale simulation benchmarks, including CALVIN, MetaWorld, and Robomimic. In CALVIN (10% data split), PRISM improves success rates by approximately 25% over diffusion and approximately 20% over flow matching, while simultaneously reducing trajectory jerk by 20x-50x. These results position PRISM as a fast, accurate, and multisensory imitation policy that retains multimodal action coverage without the latency of iterative sampling.

Learning-based Initialization of Trajectory Optimization for Path-following Problems of Redundant Manipulators

Trajectory optimization (TO) is an efficient tool to generate a redundant manipulator's joint trajectory following a 6-dimensional Cartesian path. The optimization performance largely depends on the quality of initial trajectories. However, the selection of a high-quality initial trajectory is non-trivial and requires a considerable time budget due to the extremely large space of the solution trajectories and the lack of prior knowledge about task constraints in configuration space. To alleviate the issue, we present a learning-based initial trajectory generation method that generates high-quality initial trajectories in a short time budget by adopting example-guided reinforcement learning. In addition, we suggest a null-space projected imitation reward to consider null-space constraints by efficiently learning kinematically feasible motion captured in expert demonstrations. Our statistical evaluation in simulation shows the improved optimality, efficiency, and applicability of TO when we plug in our method's output, compared with three other baselines. We also show the performance improvement and feasibility via real-world experiments with a seven-degree-of-freedom manipulator.

Towards Exploratory and Focused Manipulation with Bimanual Active Perception: A New Problem, Benchmark and Strategy

Recently, active vision has reemerged as an important concept for manipulation, since visual occlusion occurs more frequently when main cameras are mounted on the robot heads. We reflect on the visual occlusion issue and identify its essence as the absence of information useful for task completion. Inspired by this, we come up with the more fundamental problem of Exploratory and Focused Manipulation (EFM). The proposed problem is about actively collecting information to complete challenging manipulation tasks that require exploration or focus. As an initial attempt to address this problem, we establish the EFM-10 benchmark that consists of 4 categories of tasks that align with our definition (10 tasks in total). We further come up with a Bimanual Active Perception (BAP) strategy, which leverages one arm to provide active vision and another arm to provide force sensing while manipulating. Based on this idea, we collect a dataset named BAPData for the tasks in EFM-10. With the dataset, we successfully verify the effectiveness of the BAP strategy in an imitation learning manner. We hope that the EFM-10 benchmark along with the BAP strategy can become a cornerstone that facilitates future research towards this direction. Project website: EFManipulation.github.io.

Didactic to Constructive: Turning Expert Solutions into Learnable Reasoning

Improving the reasoning capabilities of large language models (LLMs) typically relies either on the model's ability to sample a correct solution to be reinforced or on the existence of a stronger model able to solve the problem. However, many difficult problems remain intractable for even current frontier models, preventing the extraction of valid training signals. A promising alternative is to leverage high-quality expert human solutions, yet naive imitation of this data fails because it is fundamentally out of distribution: expert solutions are typically didactic, containing implicit reasoning gaps intended for human readers rather than computational models. Furthermore, high-quality expert solutions are expensive, necessitating generalizable sample-efficient training methods. We propose Distribution Aligned Imitation Learning (DAIL), a two-step method that bridges the distributional gap by first transforming expert solutions into detailed, in-distribution reasoning traces and then applying a contrastive objective to focus learning on expert insights and methodologies. We find that DAIL can leverage fewer than 1000 high-quality expert solutions to achieve 10-25% pass@k gains on Qwen2.5-Instruct and Qwen3 models, improve reasoning efficiency by 2x to 4x, and enable out-of-domain generalization.

TIC-VLA: A Think-in-Control Vision-Language-Action Model for Robot Navigation in Dynamic Environments

Robots in dynamic, human-centric environments must follow language instructions while maintaining real-time reactive control. Vision-language-action (VLA) models offer a promising framework, but they assume temporally aligned reasoning and control, despite semantic inference being inherently delayed relative to real-time action. We introduce Think-in-Control (TIC)-VLA, a latency-aware framework that explicitly models delayed semantic reasoning during action generation. TIC-VLA defines a delayed semantic-control interface that conditions action generation on delayed vision-language semantic states and explicit latency metadata, in addition to current observations, enabling policies to compensate for asynchronous reasoning. We further propose a latency-consistent training pipeline that injects reasoning inference delays during imitation learning and online reinforcement learning, aligning training with asynchronous deployment. To support realistic evaluation, we present DynaNav, a physics-accurate, photo-realistic simulation suite for language-guided navigation in dynamic environments. Extensive experiments in simulation and on a real robot show that TIC-VLA consistently outperforms prior VLA models while maintaining robust real-time control under multi-second reasoning latency. Project website: https://ucla-mobility.github.io/TIC-VLA/

ForSim: Stepwise Forward Simulation for Traffic Policy Fine-Tuning

As the foundation of closed-loop training and evaluation in autonomous driving, traffic simulation still faces two fundamental challenges: covariate shift introduced by open-loop imitation learning and limited capacity to reflect the multimodal behaviors observed in real-world traffic. Although recent frameworks such as RIFT have partially addressed these issues through group-relative optimization, their forward simulation procedures remain largely non-reactive, leading to unrealistic agent interactions within the virtual domain and ultimately limiting simulation fidelity. To address these issues, we propose ForSim, a stepwise closed-loop forward simulation paradigm. At each virtual timestep, the traffic agent propagates the virtual candidate trajectory that best spatiotemporally matches the reference trajectory through physically grounded motion dynamics, thereby preserving multimodal behavioral diversity while ensuring intra-modality consistency. Other agents are updated with stepwise predictions, yielding coherent and interaction-aware evolution. When incorporated into the RIFT traffic simulation framework, ForSim operates in conjunction with group-relative optimization to fine-tune traffic policy. Extensive experiments confirm that this integration consistently improves safety while maintaining efficiency, realism, and comfort. These results underscore the importance of modeling closed-loop multimodal interactions within forward simulation and enhance the fidelity and reliability of traffic simulation for autonomous driving. Project Page: https://currychen77.github.io/ForSim/

HumanX: Toward Agile and Generalizable Humanoid Interaction Skills from Human Videos

Enabling humanoid robots to perform agile and adaptive interactive tasks has long been a core challenge in robotics. Current approaches are bottlenecked by either the scarcity of realistic interaction data or the need for meticulous, task-specific reward engineering, which limits their scalability. To narrow this gap, we present HumanX, a full-stack framework that compiles human video into generalizable, real-world interaction skills for humanoids, without task-specific rewards. HumanX integrates two co-designed components: XGen, a data generation pipeline that synthesizes diverse and physically plausible robot interaction data from video while supporting scalable data augmentation; and XMimic, a unified imitation learning framework that learns generalizable interaction skills. Evaluated across five distinct domains--basketball, football, badminton, cargo pickup, and reactive fighting--HumanX successfully acquires 10 different skills and transfers them zero-shot to a physical Unitree G1 humanoid. The learned capabilities include complex maneuvers such as pump-fake turnaround fadeaway jumpshots without any external perception, as well as interactive tasks like sustained human-robot passing sequences over 10 consecutive cycles--learned from a single video demonstration. Our experiments show that HumanX achieves over 8 times higher generalization success than prior methods, demonstrating a scalable and task-agnostic pathway for learning versatile, real-world robot interactive skills.