The AI for Industry Challenge targets a critical bottleneck in modern manufacturing: electronics assembly. Specifically, it focuses on dexterous cable management and insertion—a task that currently remains largely manual and repetitive.
From a robotics perspective, this task is notoriously difficult due to the complex physics involved in manipulating flexible cables and the extreme precision required to perceive, handle, and insert connectors.
The Goal: Participants will train an AI model using open-source simulators (e.g., Isaac Sim, MuJoCo, Gazebo), leveraging ROS for communication. This is your opportunity to bridge the sim-to-real gap and make tangible progress on a significant real-world problem.
The Reward: Finalists will deploy their models from simulation to a physical workcell hosted at Intrinsic’s HQ. The top five teams will share a $180,000 prize pool.
The challenge officially begins on March 2 and runs through September 8, 2026. It consists of three distinct phases:
- Qualification (Mar 2 - May 15): Participants train and test their cable assembly models in simulation. Evaluation Period: May 18 - 27. Top 30 announced May 28.
- Phase 1 (May 28 - Jul 14): Qualified teams advance and gain access to Intrinsic Flowstate to develop a complete cable handling solution. Evaluation Period: Jul 14 - 21. Top 10 announced Jul 22.
- Phase 2 (Jul 27 - Aug 25): Top teams move on to deploy and refine their solutions on a physical workcell provided by Intrinsic for real-world testing and evaluation. Evaluation Period: Aug 26 - Sep 4. Winner announced Sep 8.
For a detailed breakdown of expectations and deliverables, please refer to Competition Phases.
Scoring across all three phases is automated. Rankings are determined by a combination of the following criteria:
- Model Validity: Submissions must load without errors and generate valid robot commands on the required ROS topics. Invalid submissions are disqualified.
- Task Success: A binary metric applied to each successful cable insertion.
- Precision: Scores based on how closely the connectors are inserted relative to their target pose.
- Safety: Penalties applied for collisions or excessive force exerted on connectors or cables.
- Efficiency: Measurement of the overall cycle time to complete the assembly tasks; faster solutions are rewarded.
For details, see Scoring and the Scoring Test & Evaluation Guide.
To advance in the challenge and remain eligible for prizes, teams must submit their models at the end of each phase.
- Authentication: Each team leader will receive a unique authentication token for uploads.
- Frequency: Teams may make multiple submissions prior to the deadline; the final submission will be used for scoring.
For detailed upload instructions, see Submission Guidelines.
We provide several baseline policy implementations to help you get started, including a minimal example, a ground truth-based policy for debugging, and an ACT (Action Chunking with Transformers) policy.
For details on running these policies, see the Example Policies README and the Policy Integration Guide.
Ready to begin? Please consult the Getting Started Guide.