Robot model predicts motion and stiffness for delicate tasks
CompVLA: A Variable Compliance Vision-Language-Action Model for Contact-rich Manipulation
Robotics
Summary
Robots doing tasks that involve touching and pushing objects need to control both how they move and how much they resist forces. Existing robot systems mostly just decide how to move, which can cause problems with delicate tasks. The authors created CompVLA, a robot model that uses pictures and language instructions to decide both motion and flexibility. This lets the robot handle tasks better and more successfully than previous methods.
What this means in practice
- •For industrial robot developers: Enable robots to better handle tasks requiring gentle and adaptive force control guided by vision and language inputs.$Commercial implications: Allows development of robots for manufacturing or assembly lines that perform delicate contact-rich manipulations reliably.
- •For robotics software engineers: Build robot control software that integrates compliance parameters with motion commands for improved real-world manipulation.
Authors
Jongmin Kim, Junsu Ha, Che-Sang Park, Minchang Song, Hyeokju Jeong, Himchan Hwang, Jianlong Fu, Frank C. Park
Abstract
Contact-rich manipulation, requiring robots to regulate not only motion but also how they yield to external forces, has emerged as the next frontier for Vision-Language-Action (VLA) models. However, existing VLAs output purely kinematic commands, degrading performance on real-world contact-rich tasks. In this paper, we introduce CompVLA, a unified VLA framework that jointly predicts motion and stiffness matrix from RGB and language inputs. Our approach augments the conventional architecture with a dedicated Compliance Expert, which outputs time-varying stiffness and virtual displacement profiles executed via geometric impedance control. We demonstrate that CompVLA achieves the highest average success rate across diverse contact-rich tasks, outperforming both vanilla and compliance-aware VLA baselines, with ablations confirming each component is essential.