# Learning Precise, Contact-Rich Manipulation through Uncalibrated Tactile Skins

**Authors:** Venkatesh Pattabiraman, Yifeng Cao, Siddhant Haldar, Lerrel Pinto, Raunaq Bhirangi  
**arXiv:** [2410.17246](https://arxiv.org/abs/2410.17246)  
**Date:** 2024-10-22  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2410.17246v1/

## Key points

- Authors Venkatesh Pattabiraman, Yifeng Cao, Siddhant Haldar, Lerrel Pinto, and Raunaq Bhirangi introduced a groundbreaking approach in robotic manipulation with their work on Learning Precise, Contact-Rich Manipulation through Uncalibrated Tactile Skins.
- The research addresses challenges faced in executing contact-rich tasks accurately due to limitations of vision in understanding physical interactions.
- Integrating tactile sensing into policy learning is essential for enhancing performance in robotic manipulation.
- The team's work explores the use of magnetic skin sensors as a solution to this challenge due to their advantages such as being low-dimensional, highly sensitive, and cost-effective.
- ViSk framework utilizes a transformer-based policy that incorporates skin sensor data alongside visual information for improved performance.
- Experiments conducted on four complex real-world tasks show that ViSk outperforms vision-only policies and those based on optical tactile sensing.
- Combining tactile and visual modalities leads to enhanced policy performance and spatial generalization in robotic manipulation tasks.

## Summary

Authors Venkatesh Pattabiraman, Yifeng Cao, Siddhant Haldar, Lerrel Pinto, and Raunaq Bhirangi have introduced a groundbreaking approach in robotic manipulation with their work on Learning Precise, Contact-Rich Manipulation through Uncalibrated Tactile Skins. The research addresses the challenges faced in executing contact-rich tasks accurately due to the limitations of vision in understanding physical interactions. While visuomotor policy learning has made significant advancements in robotic manipulation, integrating tactile sensing into policy learning has become essential for enhancing performance. The team's work explores the use of magnetic skin sensors as a solution to this challenge. These sensors offer inherent advantages such as being low-dimensional, highly sensitive, and cost-effective to integrate with robotic platforms. is an ever-evolving field that constantly seeks new ways to improve efficiency and precision. In this regard,  plays a crucial role in enabling robots to interact with their environment effectively. To effectively leverage these sensors, they introduce the  This framework utilizes a transformer-based policy that incorporates skin sensor data as additional tokens alongside visual information. To evaluate the effectiveness of their approach, the researchers conducted experiments on four complex real-world tasks: credit card swiping, plug insertion, USB insertion, and bookshelf retrieval. The results demonstrate that ViSk outperforms both vision-only policies and those based on optical tactile sensing. Further analysis reveals that combining tactile and visual modalities leads to enhanced policy performance and spatial generalization,

Overall, by demonstrating how integrating tactile sensing can improve the execution of contact-rich tasks. The  offers a simple yet powerful approach that showcases the potential of  in enhancing robotic capabilities. Through their innovative work, the authors have paved the way for more efficient and precise manipulation tasks in robotics using uncalibrated tactile skins.