>
Breaking Lock-In: Preserving Steerability under Low-Data VLA Post-Training
Mitigating lock-in in low-data VLA post-training by preserving visual grounding and using test-time contrastive prompt guidance to improve instruction steerability.
LEGS: Fine-Tuning Teleop-Free VLAs for Humanoid Loco-manipulation in an Embodied Gaussian Splatting World
LEGS composites the robot and objects over a photorealistic 3D Gaussian Splatting background inside MuJoCo, and procedurally generates labeled demonstrations — no teleoperation, no seed demos, no human video.
Ego-Pi: VLA Fine-Tuning for Ego-Centric Human and Robot Data
Fine-tuning vision-language-action (VLA) models for dexterous humanoid manipulation using aligned egocentric human demonstrations and robot data.
Bionic lightweight design of limb leg units for hydraulic quadruped robots by additive manufacturing and topology optimization
We design a bionic lightweight limb leg unit for hydraulic quadruped using topology optimization and additive manufacturing, reduced weight and improved performance.