Advances in Legged Robotics and Manipulation

The field of robotics is witnessing significant advancements in legged robotics and manipulation, with a focus on developing more agile, adaptive, and robust systems. Recent research has explored the use of learning-based approaches, such as reinforcement learning and imitation learning, to improve the control and manipulation capabilities of legged robots. These approaches have shown promise in enabling robots to learn complex tasks, such as locomotion, grasping, and manipulation, in a more efficient and effective manner. Notably, the development of hierarchical control frameworks and the integration of tactile sensing and feedback have also been key areas of research, allowing for more precise and adaptive control of robotic systems. Furthermore, the application of these technologies to real-world scenarios, such as search and rescue, and logistics, is becoming increasingly prominent. Overall, the field is moving towards the development of more autonomous, flexible, and human-like robotic systems. Noteworthy papers include: Autonomous UAV-Quadruped Docking in Complex Terrains via Active Posture Alignment and Constraint-Aware Control, which proposes a framework for autonomous docking between UAVs and quadruped robots in complex environments. Learning-Based Collaborative Control for Bi-Manual Tactile-Reactive Grasping, which presents a learning-based approach for collaborative grasping and manipulation using tactile feedback. LocoFormer: Generalist Locomotion via Long-context Adaptation, which introduces a generalist locomotion model that can adapt to different robotic systems and environments.

Sources

Autonomous UAV-Quadruped Docking in Complex Terrains via Active Posture Alignment and Constraint-Aware Control

Learning Multi-Skill Legged Locomotion Using Conditional Adversarial Motion Priors

Context and Diversity Matter: The Emergence of In-Context Learning in World Models

Learning-Based Collaborative Control for Bi-Manual Tactile-Reactive Grasping

HeLoM: Hierarchical Learning for Whole-Body Loco-Manipulation in Hexapod Robot

DA-MMP: Learning Coordinated and Accurate Throwing with Dynamics-Aware Motion Manifold Primitives

LocoFormer: Generalist Locomotion via Long-context Adaptation

Unlocking the Potential of Soft Actor-Critic for Imitation Learning

JuggleRL: Mastering Ball Juggling with a Quadrotor via Deep Reinforcement Learning

TacRefineNet: Tactile-Only Grasp Refinement Between Arbitrary In-Hand Object Poses

SAC Flow: Sample-Efficient Reinforcement Learning of Flow-Based Policies via Velocity-Reparameterized Sequential Modeling

Noise-Guided Transport for Imitation Learning

DiSA-IQL: Offline Reinforcement Learning for Robust Soft Robot Control under Distribution Shifts

Shared Object Manipulation with a Team of Collaborative Quadrupeds

Real-Time Trajectory Generation and Hybrid Lyapunov-Based Control for Hopping Robots

SCANS: A Soft Gripper with Curvature and Spectroscopy Sensors for In-Hand Material Differentiation

Built with on top of