Safely Learning Visuo-Tactile Feedback Policies in Real For Industrial Insertion

Letian Fu, Huang Huang, Lars Berscheid, Hui Li, Ken Goldberg, Sachin Chitta

Abstract

Industrial insertion tasks are often performed repetitively with parts that are subject to tight tolerances and prone to breakage. Learning an industrial insertion policy in real is challenging as the collision between the parts and the environment can cause slippage or breakage of the part. In this paper, we present a safe self-supervised method to learn a visuo-tactile insertion policy that is robust to grasp pose variations. The method reduces human input and collisions between the part and the receptacle. The method divides the insertion task into two phases. In the first align phase, a tactile-based grasp pose estimation model is learned to align the insertion part with the receptacle. In the second insert phase, a vision-based policy is learned to guide the part into the receptacle. The robot uses force-torque sensing to achieve a safe self-supervised data collection pipeline. Physical experiments on the USB insertion task from the NIST Assembly Taskboard suggest that the resulting policies can achieve 45/45 insertion successes on 45 different initial grasp poses, improving on two baselines: (1) a behavior cloning agent trained on 50 human insertion demonstrations (1/45) and (2) an online RL policy (TD3) trained in real (0/45).

Introduction

Industrial insertion—fitting parts together with tight tolerances—is a cornerstone of manufacturing. However, it remains a "holy grail" for robotics due to three primary challenges. First, Grasp Uncertainty: when a robot picks up a part, the exact orientation in the gripper is often unknown and visually occluded. Second, Fragility: parts and receptacles are prone to breakage, and traditional trial-and-error Reinforcement Learning (RL) often results in high-force collisions. Finally, the Sim-to-Real Gap makes it difficult to simulate the complex physics of soft tactile sensors, necessitating real-world data collection.

This research presents a system capable of learning high-precision insertion policies directly in the real world, safely, and without requiring thousands of expensive human demonstrations.

Methodology

The proposed architecture decouples the insertion task into two distinct, manageable phases:

Phase 1: Tactile Guided Alignment ($\pi_{tac}$)

The robot utilizes a DIGIT tactile sensor—a high-resolution optical-based sensor—to "feel" the part. A neural network (RegNet) estimates the grasp pose from this tactile image. The robot then reorients the part to align its axis perfectly with the receptacle, compensating for initial grasp errors.

The two-phase approach begins with an initial grasp followed by tactile-guided alignment to correct pose offsets.

Phase 2: Vision Guided Insertion ($\pi_{vis}$)

Once aligned, a wrist-mounted RGB camera takes over. A second policy uses visual servoing to compensate for remaining $x, y$ offsets and guides the part into the hole.

Safe Self-Supervised Data Collection

To collect data safely without a human in the loop, the system employs force-torque sensing. Starting from a single imperfect human demonstration, the robot performs a "grid search" using force-torque feedback to find a "reversible" pose—a position where the part can be inserted and removed with minimal resistance. Once this "sweet spot" is identified, the robot autonomously plugs and unplugs the part thousands of times to collect training data, automatically resetting if it detects a high-force collision.

Experimental Setup

The study utilizes the NIST Assembly Taskboard, focusing on the USB-A insertion task. This task is characterized by notoriously tight tolerances that demand high precision.

  • Tactile Dataset: 2,000 data points collected in 120 minutes, mapping tactile images to 3-DoF grasp poses.
  • Vision Dataset: ~46,000 images collected across 125 different grasps to train the visual servoing policy.

Results and Discussion

The proposed method was evaluated against 45 different initial grasp poses and compared to standard baselines. The results demonstrate the superiority of the visuo-tactile approach:

  • Proposed Approach: 45/45 successes (100%).
  • Behavior Cloning (Imitation Learning): 1/45 successes. 50 human demonstrations were insufficient to cover the variety of possible grasp offsets.
  • Online RL (TD3): 0/45 successes. The policy failed to converge because frequent collisions caused the part to slip in the gripper, constantly shifting the ground truth.
Successful insertion of a USB-A connector into a receptacle using the learned visuo-tactile policy.

The tactile alignment policy proved remarkably accurate, achieving a mean error of only ~0.09mm in translation and ~0.005 radians in rotation. Key to this success was the Asymmetric Gripper Design, which used a tactile sensor on one jaw and high-friction sandpaper on the other to balance sensing resolution with grip stability.

Conclusion

By separating the problem into alignment (tactile) and insertion (vision), this method transforms a complex RL problem into simpler supervised learning tasks. While highly successful for USB connectors, future work will focus on generalizing these policies to different parts (e.g., HDMI, Ethernet) and increasing the speed of the initial "reversible pose" search to facilitate more rapid industrial deployment.

Summary updated

Abstract

Industrial insertion tasks are often performed repetitively with parts that are subject to tight tolerances and prone to breakage. Learning an industrial insertion policy in real is challenging as the collision between the parts and the environment can cause slippage or breakage of the part. In this paper, we present a safe self-supervised method to learn a visuo-tactile insertion policy that is robust to grasp pose variations. The method reduces human input and collisions between the part and the receptacle. The method divides the insertion task into two phases. In the first align phase, a tactile-based grasp pose estimation model is learned to align the insertion part with the receptacle. In the second insert phase, a vision-based policy is learned to guide the part into the receptacle. The robot uses force-torque sensing to achieve a safe self-supervised data collection pipeline. Physical experiments on the USB insertion task from the NIST Assembly Taskboard suggest that the resulting policies can achieve 45/45 insertion successes on 45 different initial grasp poses, improving on two baselines: (1) a behavior cloning agent trained on 50 human insertion demonstrations (1/45) and (2) an online RL policy (TD3) trained in real (0/45).
BibTeX
@inproceedings{fu2023safely,
  author = {Letian Fu and Huang Huang and Lars Berscheid and Hui Li and Ken Goldberg and Sachin Chitta},
  title = {{Safely Learning Visuo-Tactile Feedback Policies in Real For Industrial Insertion}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2023},
  url = {https://openreview.net/forum?id=T99R7KVe6M},
}
Figure