Robotic assistants in long-term human-robot collaboration need to assist users under partial observations while leveraging cross-day interaction history. Since human traits are often unknown initially, passive infer-then-act is ineffective. We propose PACT, an ask-or-act framework that evaluates contextual sufficiency to decide whether to seek clarification before acting. Using reinforcement learning, PACT improves assistance accuracy and clarification utility over passive baselines in multi-day embodied scenarios.

This paper presents IsaacIPC, a robotic simulation framework coupling GPU-accelerated Incremental Potential Contact (IPC) with IsaacSim/Lab. It maps simulated deformation between simulation and visual meshes for real-time realistic rendering, aiding data collection and policy evaluation. It also introduces the Geometric Mortar Contact Potential (GMCP) for improved tactile sensing contact-pressure resolution. Evaluated on contact benchmarks and demonstrated on rigid-deformable simulations including a quadruped robot, dexterous hand, and UMI gripper.

Planetary rovers face mobility challenges on varying terrains. Researchers introduce a multimodal wheel that continuously adjusts grouser height. In 750 trials across four surfaces, adaptive deployment reduced slip by 30-58% and improved travel time and energy by up to 77.4% on granular terrains, highlighting limitations of fixed wheels.

This paper proposes a reinforcement learning framework that modulates a constant reference trajectory to perform compact, position-constrained quadrotor inversions while remaining compatible with traditional trajectory generation and tracking. In simulation, the method reduces position RMSE by 32% and settling time by 57% relative to the strongest optimization-based baseline. Hardware experiments demonstrate successful inversion across multiple yaw configurations with position RMSE below 0.35m.

A new approach combines evolution and learning in robots using a mixture of experts conditioned on embodiment, balancing efficiency and adaptability without monolithic controllers.