PopcornFT 🍿💪🔧: A Compact LED-Based Displacement Sensor for Robot Fingers

🌐 Project Website

Conducted in ROAM Lab, PopcornFT is a compact, LED-based displacement sensor designed to measure forces and torques along the x, y, and z axes. The integration of sensor hardware and software includes:

• 3D-printed PLA flexures meticulously designed based on material strength
• Compact sensor circuit board, consisting of four LED emitters along with 24 LED receivers (12 on both the upper and lower surfaces), enabling it to generate signals highly sensitive to displacement.
• A heuristic Gaussian classification model for the contact detection
• A transformer-based regression model, which determines contact events and estimates the applied force/torque vectors in 3D, while maintaining strong dynamic robustness

My contributions to this project include：

1. Data collection
2. Development and parameter tuning of the Gaussian model using grid search to improve contact detection accuracy
3. Training and hyperparameter tuning of the transformer-based regression model
4. Ablation studies for the model performance by using less LED singal for training

Contact Detection using Gaussian Model with ynamical Robustness