Challenge Overview
The TriFusion Challenge addresses two core sim-to-real problems at the convergence of computer graphics and robotics: universal motion tracking for full-scale humanoids, and visuo-tactile simulation optimization for dexterous manipulation.
Track 1 focuses on high-fidelity motion retargeting from human performance to simulated humanoids, evaluated on precise positional and DOF tracking metrics.
Track 2 targets improving the realism and accuracy of visuo-tactile rendering and physical modeling in peg-in-hole tasks using RoboTwin and GelSight Mini data.
Competition Tracks
🎯 Track 1: Universal Motion Tracking
Objective: High-fidelity motion retargeting from human performance to simulated humanoids with precise tracking evaluation.
Key Tasks:
- Full-scale humanoid motion capture
- Real-time motion retargeting
- DOF (Degrees of Freedom) optimization
- Cross-morphology adaptation
Evaluation Metrics:
- Positional tracking accuracy
- DOF tracking precision
- Motion fidelity scores
- Temporal consistency
🔬 Track 2: Visuo-Tactile Simulation
Objective: Improve realism and accuracy of visuo-tactile rendering and physical modeling for dexterous manipulation tasks.
Key Tasks:
- Peg-in-hole task optimization
- RoboTwin data integration
- GelSight Mini sensor modeling
- Haptic feedback simulation
Evaluation Metrics:
- Tactile rendering accuracy
- Physical model fidelity
- Task completion rates
- Sim-to-real transfer success
Challenge Timeline
Important Dates
- Challenge Launch: September 1, 2025
- Registration Deadline: October 15, 2025
- Dataset Release: September 15, 2025
- Submission Deadline: November 20, 2025
- Results Announcement: December 1, 2025
- Workshop Presentation: December 8, 2025
Participation Guidelines
Eligibility
- Open to academic institutions, research labs, and industry teams
- Teams can have 2-8 members
- Multiple submissions per team allowed across different tracks
Submission Requirements
- Technical Report: 4-8 pages describing methodology and results
- Code: Complete implementation with documentation
- Demo Video: 3-5 minutes showcasing your solution
- Reproducibility Package: Data, models, and evaluation scripts
Evaluation Process
- Automated Evaluation: Quantitative metrics on benchmark datasets
- Human Evaluation: Qualitative assessment by expert judges
- Live Demonstration: Real-time system performance at the workshop
- Peer Review: Technical merit assessment by the community
Prizes and Recognition
🥇 Best Overall Solution
$5,000 USD
Awarded to the team with the most innovative and effective solution across all evaluation criteria.
🏆 Track Winners
$2,000 USD each
Best solution in each of the two competition tracks (Universal Motion Tracking, Visuo-Tactile Simulation).
🌟 Innovation Award
$1,000 USD
Most creative and novel approach, as judged by the technical committee.
Resources and Support
Datasets
- TriFusion Benchmark: Comprehensive multi-modal dataset for embodied intelligence
- Simulation Environments: Unity and PyBullet environments for virtual testing
- Hardware Access: Limited remote access to humanoid robot platforms
Technical Support
- Discord Community: Real-time discussion and Q&A
- Office Hours: Weekly video calls with organizers
- Documentation: Comprehensive API documentation and tutorials
Registration
Ready to push the boundaries of embodied intelligence?
Register Your TeamQuestions?
For more information about the TriFusion Challenge, please contact us at trifusion2025@gmail.com or join our community discussions.