In many robotic systems where dynamics are best modeled as stochastic systems due to factors such as sensing noise and environmental disturbances, it is challenging for conventional methods such as Hamilton-Jacobi reachability and control barrier functions to provide a holistic measure of safety. We derive a linear operator governing the dynamic programming principle for safety probability, and find that its dominant eigenpair provides information about safety for both individual states and the overall closed-loop system. The proposed learning framework, called EigenSafe, jointly learns this dominant eigenpair and a safe backup policy in an offline manner. The learned eigenfunction is then used to construct a safety filter that detects potentially unsafe situations and falls back to the backup policy. The framework is validated in three simulated stochastic safety-critical control tasks.

Reinforcement learning (RL) has become a standard paradigm for refining large language models (LLMs) beyond pre-training and instruction tuning. A prominent line of work is RL with verifiable rewards (RLVR), which leverages automatically verifiable outcomes (e.g., correctness or executability) to generate reward signals. While efficient, this framework faces two key limitations: First, its binary feedback is too sparse to capture the quality of the reasoning process. Second, its coarse-grained rewards potentially lead to vanishing gradients. Inspired by observations from human learning, we introduce a RL technique that integrates verifiable outcomes with the model's own confidence estimates. This joint design enriches the reward signal, providing finer-grained feedback and implicitly supervising the reasoning process. Experimental results demonstrate that our proposed method enhances RL performance across multiple datasets and reduces token consumption during inference, while incurring negligible additional training cost. Moreover, it can be used as a plug-in module to enhance other state-of-the-art RL methods.

Construction sites frequently require removing large rocks before excavation or grading can proceed. Human operators typically extract these boulders using only standard digging buckets, avoiding time-consuming tool changes to specialized grippers. This task demands manipulating irregular objects with unknown geometries in harsh outdoor environments where dust, variable lighting, and occlusions hinder perception. The excavator must adapt to varying soil resistance--dragging along hard-packed surfaces or penetrating soft ground--while coordinating multiple hydraulic joints to secure rocks using a shovel. Current autonomous excavation focuses on continuous media (soil, gravel) or uses specialized grippers with detailed geometric planning for discrete objects. These approaches either cannot handle large irregular rocks or require impractical tool changes that interrupt workflow. We train a reinforcement learning policy in simulation using rigid-body dynamics and analytical soil models. The policy processes sparse LiDAR points (just 20 per rock) from vision-based segmentation and proprioceptive feedback to control standard excavator buckets. The learned agent discovers different strategies based on soil resistance: dragging along the surface in hard soil and penetrating directly in soft conditions. Field tests on a 12-ton excavator achieved 70% success across varied rocks (0.4-0.7m) and soil types, compared to 83% for human operators. This demonstrates that standard construction equipment can learn complex manipulation despite sparse perception and challenging outdoor conditions.

Abstract-- Object insertion tasks are prone to failures under pose uncertainties and environmental variations, traditionally requiring manual finetuning or controller retraining. We present a novel approach for robust and resilient object insertion using a passively compliant soft wrist that enables safe contact absorption through large deformations, without high-frequency control or force sensing. Our method structures insertion as compliance-enabled contact formations, sequential contact states that progressively constrain degrees of freedom, and integrates automated failure recovery strategies. Our key insight is that wrist compliance permits safe, repeated recovery attempts; hence, we refer to it as compliance-enabled failure recovery. We employ a pre-trained vision-language model (VLM) that assesses each skill execution from terminal poses and images, identifies failure modes, and proposes recovery actions by selecting skills and updating goals. I. INTRODUCTION Peg-in-hole tasks have been widely studied but remain challenging due to their contact-rich nature and tight tolerances [1]. A central difficulty lies in handling uncertainties in part grasping and hole pose. Conventional methods address these uncertainties through precise pose estimation and rigid fixturing, enabling repetitive pick-and-place operations [2], but such approaches demand substantial engineering effort.

Abstract-- The complex nonlinear dynamics of hydraulic excavators, such as time delays and control coupling, pose significant challenges to achieving high-precision trajectory tracking. Traditional control methods often fall short in such applications due to their inability to effectively handle these nonlinearities, while commonly used learning-based methods require extensive interactions with the environment, leading to inefficiency. T o address these issues, we introduce EfficientTrack, a trajectory tracking method that integrates model-based learning to manage nonlinear dynamics and leverages closed-loop dynamics to improve learning efficiency, ultimately minimizing tracking errors. Comparative experiments in simulation demonstrate that our method outperforms existing learning-based approaches, achieving the highest tracking precision and smoothness with the fewest interactions. Real-world experiments further show that our method remains effective under load conditions and possesses the ability for continual learning, highlighting its practical applicability. Excavators are primarily used in earthworks, mining, and construction projects, playing a vital role in tasks such as digging, loading, trenching, and leveling [1], [2], [3].

Generating obstacle-free trajectories for robotic manipulators in unstructured and cluttered environments remains a significant challenge. Existing motion planning methods often require additional computational effort to generate the final trajectory by solving kinematic or dynamic equations. This paper highlights the strong potential of model-free reinforcement learning methods over model-based approaches for obstacle-free trajectory planning in joint space. We propose a fast trajectory planning system for manipulators that combines vision-based path planning in task space with reinforcement learning-based obstacle avoidance in joint space. We divide the framework into two key components. The first introduces an innovative vision-based trajectory planner in task space, leveraging the large-scale fast segment anything (FSA) model in conjunction with basis spline (B-spline)-optimized kinodynamic path searching. The second component enhances the proximal policy optimization (PPO) algorithm by integrating action ensembles (AE) and policy feedback (PF), which greatly improve precision and stability in goal-reaching and obstacle avoidance within the joint space. These PPO enhancements increase the algorithm's adaptability across diverse robotic tasks, ensuring consistent execution of commands from the first component by the manipulator, while also enhancing both obstacle avoidance efficiency and reaching accuracy. The experimental results demonstrate the effectiveness of PPO enhancements, as well as simulation-to-simulation (Sim-to-Sim) and simulation-to-reality (Sim-to-Real) transfer, in improving model robustness and planner efficiency in complex scenarios. These enhancements allow the robot to perform obstacle avoidance and real-time trajectory planning in obstructed environments. Project page available at: https://sites.google.com/view/ftp4rm/home

Contact-rich manipulation is crucial for robots to perform tasks requiring precise force control, such as insertion, assembly, and in-hand manipulation. However, most imitation learning (IL) policies remain position-centric and lack explicit force awareness, and adding force/torque sensors to collaborative robot arms is often costly and requires additional hardware design. To overcome these issues, we propose FILIC, a Force-guided Imitation Learning framework with impedance torque control. FILIC integrates a Transformer-based IL policy with an impedance controller in a dual-loop structure, enabling compliant force-informed, force-executed manipulation. For robots without force/torque sensors, we introduce a cost-effective end-effector force estimator using joint torque measurements through analytical Jacobian-based inversion while compensating with model-predicted torques from a digital twin. We also design complementary force feedback frameworks via handheld haptics and VR visualization to improve demonstration quality. Experiments show that FILIC significantly outperforms vision-only and joint-torque-based methods, achieving safer, more compliant, and adaptable contact-rich manipulation. Our code can be found in https://github.com/TATP-233/FILIC.

The advancement of foundation models fosters new initiatives for policy learning in achieving safe and efficient autonomous driving. However, a critical bottleneck lies in the manual engineering of reward functions and training curricula for complex and dynamic driving tasks, which is a labor-intensive and time-consuming process. To address this problem, we propose OGR (Orchestrate, Generate, Reflect), a novel automated driving policy learning framework that leverages vision-language model (VLM)-based multi-agent collaboration. Our framework capitalizes on advanced reasoning and multimodal understanding capabilities of VLMs to construct a hierarchical agent system. Specifically, a centralized orchestrator plans high-level training objectives, while a generation module employs a two-step analyze-then-generate process for efficient generation of reward-curriculum pairs. A reflection module then facilitates iterative optimization based on the online evaluation. Furthermore, a dedicated memory module endows the VLM agents with the capabilities of long-term memory. To enhance robustness and diversity of the generation process, we introduce a parallel generation scheme and a human-in-the-loop technique for augmentation of the reward observation space. Through efficient multi-agent cooperation and leveraging rich multimodal information, OGR enables the online evolution of reinforcement learning policies to acquire interaction-aware driving skills. Extensive experiments in the CARLA simulator demonstrate the superior performance, robust generalizability across distinct urban scenarios, and strong compatibility with various RL algorithms. Further real-world experiments highlight the practical viability and effectiveness of our framework. The source code will be available upon acceptance of the paper.

Direct Preference Optimization (DPO) is a powerful paradigm to align language models with human preferences using pairwise comparisons. However, its binary win-or-loss supervision often proves insufficient for training small models with limited capacity. Prior works attempt to distill information from large teacher models using behavior cloning or KL divergence. These methods often focus on mimicking current behavior and overlook distilling reward modeling. To address this issue, we propose \textit{Teacher Value-based Knowledge Distillation} (TVKD), which introduces an auxiliary reward from the value function of the teacher model to provide a soft guide. This auxiliary reward is formulated to satisfy potential-based reward shaping, ensuring that the global reward structure and optimal policy of DPO are preserved. TVKD can be integrated into the standard DPO training framework and does not require additional rollouts. Our experimental results show that TVKD consistently improves performance across various benchmarks and model sizes.

In audio-visual navigation (A VN) tasks, an embodied agent must autonomously localize a sound source in unknown and complex 3D environments based on audio-visual signals. Existing methods often rely on static modality fusion strategies and neglect the spatial cues embedded in stereo audio, leading to performance degradation in cluttered or occluded scenes. To address these issues, we propose an end-to-end reinforcement learning-based AVN framework with two key innovations: (1) a Stereo-Aware Attention Module (SAM), which learns and exploits the spatial disparity between left and right audio channels to enhance directional sound perception; and (2) an Audio-Guided Dynamic Fusion Module (AGDF), which dynamically adjusts the fusion ratio between visual and auditory features based on audio cues, thereby improving robustness to environmental changes. Extensive experiments are conducted on two realistic 3D scene datasets, Replica and Matterport3D, demonstrating that our method significantly outperforms existing approaches in terms of navigation success rate and path efficiency. Notably, our model achieves over 40% improvement under audio-only conditions compared to the best-performing baselines.