ManiSkill-ViTac 2025: Challenge on Manipulation Skill Learning With Vision and Tactile Sensing

News

19 Nov 2024

Official document released on arXiv! arXiv:2411.12503 🚀

16 Nov 2024

Track 2 is online! 🚀

22 OCT 2024

Track 3 is online! 🚀

9 OCT 2024

The official repository for Challenge is here. 🚀

1 OCT 2024

The challenge webpage is online! 🚀

Overview

Manipulation is the cornerstone of robotics and embodied AI. The ManiSkill-Vitac Challenge focuses on contact-rich manipulation skill learning with tactile and visual sensing and aims to provide a standardized benchmarking platform for evaluating the performance. In this challenge, we build a simulation platform equipped with high-fidelity tactile and depth sensing simulation , allowing researchers to focus their efforts on policy learning. Additionally, we've built a physical platform that aligns with the simulation. Top participants in the simulation evaluation will be further assessed on this platform to test the Sim2Real capability.

Building upon last year's ManiSkill-Vitac-Challenge-2024, we expand the discipline and design three tracks, each representing a different perspective on manipulation. We hope it will bring together researchers in tactile sensing, computer vision and policy learning, driving the development of advanced manipulation skill learning and embodied AI techniques.

The ManiSkill-ViTac Challenge 2025 features:

  • Real-world robot evaluation
  • High-fidelity tactile and depth sensing simulation
  • Contact-rich manipulation tasks that closely resemble real-world applications

Challenge Format

This challenge consists of three independent tracks. Participating teams can choose to compete in 1-3 tracks when registering. Each track is completely separate and does not interfere with the others. The evaluation criteria and advancement standards for each track are different.

Track 1: Visuotactile Manipulation

  • Input Information: Tactile information only
  • Stage 1: Ranking based on evaluation results in simulation
  • Stage 2: Only teams advancing from Stage 1 participate, submitting their best checkpoint
  • Scoring Criteria: Success rate of Task 1 * 50 + Success rate of Task 2 * 50

Track 2: Tactile-Vision-Fusion Manipulation

  • Input Information: Tactile information + depth with semantic segmentation
  • Stage 1: Ranking based on evaluation results in simulation
  • Stage 2: Only teams advancing from Stage 1 participate, submitting their best checkpoint
  • Scoring Criteria: Success rate * 100

Track 3: Sensor Structure Design

  • Design Content: Design the shape of the silicone for the GelSight Mini and the distribution of the markers
  • Evaluation Method: Success rate * 50 + Subjective score * 50, where the subjective score is rated on a scale from 0 to 1

Participation Guidelines

  • Teams can choose to participate in 1-3 tracks
  • Each track is conducted independently and does not affect the others
  • Stage 2 of Track 1 and Track 2 is only open to teams that succeed in Stage 1. The number of teams advancing to Stage 2 will be determined based on the total number of registered participants.

Challenge Process

  • During the registration phase, teams need to sign up. Please note that teams participating in Track 1 and Track 2 must register and submit their results before the Stage 1 deadline. Teams participating in Track 3 need to register and submit their results before the Track 3 deadline. Teams are allowed to participate in multiple tracks, and we recommend that teams register as early as possible to help us collect team information

Tracks

Track 1: Visuotactile Manipulation

This track includes two tasks: peg insertion and lock opening. Similar to the ManiSkill challenge, we focus on evaluating the policy’s generalizability. Each task includes multiple objects of different shapes (e.g., the peg can be a cuboid or a trapezoid, and keys can have two or three teeth), which are unknown during evaluation. Therefore, the policy needs to be able to generalize to different pegs and keys.

Peg Insertion GIF Lock Opening GIF Peg Insertion GIF

Track 2: Tactile-Vision-Fusion Manipulation

In this track, the experimental setup remains unchanged. However, compared to the only tactile task, a Realsense D415 is used to provide visual information, and the distance between the peg and the hole is increased. The gripper needs to determine which hole to insert the peg into based on the shape of the gripped peg. Since the tactile information remains the same during the approach to the hole, the task primarily relies on visual data for decision-making.

Lock Opening GIF Lock Opening GIF Peg Insertion GIF

Track 3: Sensor Structure Design

In this track, we evaluate the structural design of tactile sensors using the Gelsight Mini as a reference model. Teams are allowed to freely modify the shape of the silicone part of the sensor, as well as the number and distribution of markers. However, it is important to note that the silicone must remain firmly attached to the base, meaning the base size must stay fixed at 20.75mm x 25.25mm. The physical parameters of the silicone will be provided by us.

GIF 1
GIF 2
GIF 3
GIF 1
GIF 2
GIF 3

LeaderBoard

Track 1(Stage 1)
Team task 1 task 2 Score
# Baseline 43.89%
Track 2(Stage 1)
Team Score
# Baseline
Track 3
Team Task Innovation Score
# Baseline 43.89%

Timeline

Release

Nov 15, 2024

Preparation

Nov 15, 2024 to

Dec 31, 2024

Stage1

Jan 1, 2025 to

Feb 15, 2025 20:00

(UTC-12)

Stage2

Feb 16, 2025 to

Mar 20, 2025 20:00

(UTC-12)

Winner Announcement

Mar 29, 2025

Award Ceremony

2025 ICRA ViTac Workshop (TBD)

Award

Track 1

First Prize
$2,000 × 1
Second Prize
$1,000 × 2
Third Prize
$500 × 3

Track 2

First Prize
$2,000 × 1
Second Prize
$1,000 × 2
Third Prize
$500 × 3

Track 3

First Prize
$2,000 × 1
Second Prize
$1,000 × 1
Third Prize
$500 × 1

First prize winners of each track will be invited to give a 6-minute presentation at the ICRA 2025 ViTac Workshop.

Organizers

Faculty

Organizer 1

Rui Chen
Tsinghua University

Organizer 1

Hao Su
UC San Diego

Organizer 1

Shan Luo
King's College London

Organizer 1

Nathan Lepora
University of Bristol

Organizer 1

Roberto Calandra
TU Dresden

Organizer 1

Hamidreza Kasaei
University of Groningen

Organizer 1

Jing Xu
Tsinghua University

Student

Organizer 1

Chuanyu Li
Tsinghua University* Sichuan University

Student 1

Renjun Dang
Tsinghua University

Student 1

Xiang Li
Tsinghua University* Beijing University of Chemical Technology

Student 1

Zhiyuan Wu
King's College London

Student 1

Yuhao Liu
Tsinghua University

* work done during internship at Tsinghua University

Contact Us

Feel free to reach out to us through Discord or Gmail. We are here to assist you!

Join us on Discord

Recommended

discord-link

Email us at Gmail

maniskill.vitac@gmail.com

Sign Up for the Challenge

Register a team to make submissions.Teams shall have no more than one leader and one instructor. The maximum number of additional team members is four. Fill in the form below to register for the challenge. We will contact you with further details.