Advances in Robotic Manipulation and Locomotion

The field of robotics is witnessing significant advancements in manipulation and locomotion capabilities. Researchers are exploring innovative approaches to enable robots to learn and adapt to new tasks and environments, with a focus on developing more robust and generalizable policies. One of the key trends is the use of diffusion-based models and reinforcement learning to improve the efficiency and effectiveness of robotic learning. Another area of interest is the development of whole-body control frameworks for humanoid robots, which aims to enhance their balance and stability in dynamic environments. Furthermore, there is a growing emphasis on leveraging tactile modalities and multimodal sensing to improve manipulation capabilities. Noteworthy papers in this area include mimic-one, which presents a scalable model recipe for real-world control of a highly dexterous humanoid robotic hand, and SAIL, which introduces a novel approach for faster-than-demonstration execution of imitation learning policies. Additionally, papers like DynaGuide and Latent Action Diffusion demonstrate the potential of diffusion policies and latent action spaces for improving robotic manipulation and cross-embodiment learning.

Sources

Multi-Loco: Unifying Multi-Embodiment Legged Locomotion via Reinforcement Learning Augmented Diffusion

mimic-one: a Scalable Model Recipe for General Purpose Robot Dexterity

SAIL: Faster-than-Demonstration Execution of Imitation Learning Policies

DynaGuide: Steering Diffusion Polices with Active Dynamic Guidance

Whole-Body Control Framework for Humanoid Robots with Heavy Limbs: A Model-Based Approach

Steering Robots with Inference-Time Interactions

Single-Example Learning in a Mixture of GPDMs with Latent Geometries

Latent Action Diffusion for Cross-Embodiment Manipulation

GMT: General Motion Tracking for Humanoid Whole-Body Control

Booster Gym: An End-to-End Reinforcement Learning Framework for Humanoid Robot Locomotion

TACT: Humanoid Whole-body Contact Manipulation through Deep Imitation Learning with Tactile Modality

Robust Instant Policy: Leveraging Student's t-Regression Model for Robust In-context Imitation Learning of Robot Manipulation

Learning Task-Agnostic Skill Bases to Uncover Motor Primitives in Animal Behaviors

Built with on top of