Humanoid Papers Roundup: Sit-to-Stand, Physics Filtering, and Visuomotor Skills
carlosdponx · x · 2026-09-02
A roundup of humanoid robotics papers from late August, covering locomotion, motion tracking, and loco-manipulation.
Locomotion:
- LAC: Teaches a G1 humanoid to stand from chairs of varying heights using guided assistance curricula and staged rewards.
- Physics Filtering: Filters learned controllers through simple physics to help legged robots generalize to new terrains and payloads.
- CARO: Adds a torque-residual body cue for zero-shot robust quadruped locomotion amid payload and terrain changes.
Motion Tracking:
- Demonstration-Guided: Adapts stand-up demonstrations for G1 to handle soft, compliant surfaces.
Loco-Manipulation:
- DreamMimic: Learns visuomotor whole-body loco-manipulation via distillation from simulated teachers using world models.
- TONAV: Task-oriented navigation and action-velocity chunk learning for quadrupedal manipulation of articulated objects.
Whole-Body Control: Covers MPC, impedance control, and safety layers.
Related event: Humanoid Robot Paper Digest Highlights Sit-Stand Control(2 posts)→
More from Embodied
- An AI Toothbrush Product Launched — Saboo_Shubham_ · 2026-09-02
- Is a used RTX 3060 12GB worth it for local AI video/image generation? — yeah280 · 2026-09-02
- Hardware insight: Voltage choice significantly affects motor torque limits — philfung · 2026-09-02
- Nvidia predicts physical AI could be 10x larger than digital AI — Rewkang · 2026-09-02
- ESP32 Multi-Sensor Environmental Monitor with Home Assistant — JeremyCMorgan · 2026-09-02
- SnakeROS: A Pure-Python ROS 2 Client for MicroPython — kscottz · 2026-09-02