Skip to content

Event

Tactile Reinforcement Learning at the Institute of Automation, Chinese Academy of Sciences Improves Sub-Millimeter Assembly Success Rate

3 reports · 3 sources

The team led by Cheng Long at the Institute of Automation, Chinese Academy of Sciences published research at IROS 2026 proposing a tactile-enhanced full-process "grasp-insertion" learning framework that uses tactile reinforcement learning to improve the success rate of sub-millimeter precision insertion. The approach first uses Diffusion Policy imitation learning to reach the hole opening, then uses SAC reinforcement learning to complete the insertion, and handles real tactile noise through tactile baseline compensation and balanced sampling grouped by torque. Under a 0.05 mm clearance, the experimental results reported by the team show: the pure imitation learning approach relying only on vision and pose teaching achieved a success rate of 0%, the imitation learning plus reinforcement learning approach without touch achieved 52%, and after adding fingertip tactile feedback it rose to 67%, with peak contact force reduced by 60%. The experiments covered 5 types of irregular holes—square, circular, hexagonal, L-shaped, and triangular—and set three clearance gradients: 1, 0.25, and 0.05 mm; the tactile sensor used was the PaXini PX6AX GEN3, mounted on a seven-axis robotic arm to provide real-time feedback on contact status. The team states that the feasibility of sub-millimeter precision operation has been verified, but multiple industrialization thresholds still need to be crossed.

Report timeline

  1. CAS Institute of Automation and PaXini collaborative research: tactile reinforcement learning achieves sub-millimeter precision assembly

    A team led by Cheng Long at the Institute of Automation, Chinese Academy of Sciences, published research at IROS 2026 using tactile-enhanced reinforcement learning to complete sub-millimeter precision insertion, raising the success rate from 52% for a tactile-free approach to 67% under a 0.05 mm clearance.

    OFweek Robotics ·

  2. CAS Institute of Automation and Paxini collaborative study: tactile-enhanced reinforcement learning achieves sub-millimeter precision assembly

    A team led by Cheng Long at the Institute of Automation, Chinese Academy of Sciences, published research at IROS 2026 using tactile-enhanced reinforcement learning to improve the success rate and stability of sub-millimeter precision insertion. The approach first uses Diffusion Policy imitation learning to reach the hole opening, then SAC reinforcement learning to complete the insertion, and handles real-world tactile noise through tactile baseline compensation and balanced sampling grouped by torque.

    机器人daily · 微信公众号 ·

  3. CAS Institute of Automation proposes tactile-enhanced grasp-and-insertion framework, achieving 67% success rate at 0.05 mm clearance

    The team led by Cheng Long at the Institute of Automation, Chinese Academy of Sciences, proposed a tactile-enhanced full-pipeline learning framework for grasping and insertion, achieving a 67% success rate at 0.05 mm clearance and a 60% reduction in peak contact force; the paper was accepted to IROS 2026. The research uses the PaXini PX6AX GEN3 multi-dimensional tactile sensor as its core sensing hardware, integrated into a seven-axis robotic arm, covering five geometric shapes and three clearance levels of 1, 0.25, and 0.05 mm. Under the same conditions, the tactile-free IL+RL approach achieved a 52% success rate, while the pure imitation learning approach achieved 0.

    帕西尼 · 微信公众号 ·

Related topics

PaXiniGrasping and manipulationRobot models and learningData and simulationResearch papersPerception and localization

Interest trend

Current
3
Peak
4Nov 4 01:00
Past 24 hours
–

The trend compares participants observed throughout the same period, so its range may be smaller than the current interest count. Move or tap to inspect each hour; use the arrow keys on a keyboard.