Team: Team Synapse • Team #3 — GitHub: @notaroomba, @kalo08, @oxstar123

Summary: Synapse is a teleoperated surgical robot demonstrator that uses machine learning (SMVLA) to assist high-precision tasks and reduce human error in complex medical procedures.

Demo media: Local demo video: assets/demo.mp4; photos: assets/photo1.png, assets/photo2.png, assets/photo3.png

• Demonstrate how machine learning can be used in surgical teleoperation to assist complex surgeries and reduce human error.

• SCALPEL: synchronized control arm with low-latency precision engineering links for coordinated, high-accuracy teleoperation.
• Integration of SMVLA-trained policies with real-time teleoperation and an operator override/takeover toggle for assisted autonomy. ALso has a cool cardboard person that simulates an actual surgery.

• Teleoperation / Dataset capture
  • WebSocket server (main.py) receives head/hand pose and control inputs from a Meta Quest 3 client; dataset capture and teleop logs can be placed in missions/code/ and missions/wandb/.
• Training
  • Model was trained using SMVLA and reached an MVP after approximately 50 iterations.
• Inference
  • Operator can enable SCALPEL takeover (assist/autonomy) via the Meta Quest interface; runtime started with ./run_robot.sh and main.py handles inference and control.

• Implementation follows LERobot conventions and is intended to be adaptable to similar teleoperation tasks and environments.
• Flexible and scriptable; supports a Meta Quest 3 teleop client and local runtime scripts.
• Control interfaces: WebSocket-based headset client and local scripts; ./setup.sh guides hardware/port configuration (see LERobot docs for environment setup).

For example, you can provide links to:

• Local demo video: assets/demo.mp4
• URL of your dataset in Hugging Face: https://huggingface.co/datasets/NotARoomba/synapse_5
• URL of your model in Hugging Face: https://huggingface.co/NotARoomba/eval_synapse_smvla_5
• Link to a blog post describing your work:
• Repository: https://github.com/notaroomba/synapse

This is the directory tree of this repo — please place all submission materials into the missions directory with the structure below.

AMD_Robotics_Hackathon_2025_ProjectTemplate-main/
├── README.md
└── missions
    ├── code
    │   └── 
    └── wandb
        └── 
    ├── CAD
    │   └── 
    └── calibrations
        └── 
 (e.g., leader.json, follower.json)>
json)>

$ tree outputs/train/smolvla_so101_2cube_30k_steps/wandb/
outputs/train/smolvla_so101_2cube_30k_steps/wandb/
├── debug-internal.log -> run-20251029_063411-tz1cpo59/logs/debug-internal.log
├── debug.log -> run-20251029_063411-tz1cpo59/logs/debug.log
├── latest-run -> run-20251029_063411-tz1cpo59
└── run-20251029_063411-tz1cpo59
    ├── files
    │   ├── config.yaml
    │   ├── output.log
    │   ├── requirements.txt
    │   ├── wandb-metadata.json
    │   └── wandb-summary.json
    ├── logs
    │   ├── debug-core.log -> /dataset/.cache/wandb/logs/core-debug-20251029_063411.log
    │   ├── debug-internal.log
    │   └── debug.log
    ├── run-tz1cpo59.wandb
    └── tmp
        └── code
g
    ├── run-tz1cpo59.wandb
    └── tmp
        └── code
   └── code