02Jun 2024 — PresentLead researcher · IIT Madras
RL-Based Model for Assisting Stroke Patients
A reinforcement-learning agent that learns a virtual tennis game built for stroke rehabilitation — so it can anticipate a patient's movements and play alongside them, assisting recovery during therapy.
PythonDQNPeaceful PieGPU
[ gameplay capture / agent–patient rally ]
// approach
→Train a DQN agent to play the virtual tennis game developed for rehabilitation.
→Bridge the Python agent to the Unity game environment through the Peaceful Pie module.
→Have the agent anticipate the patient's returns and position itself to keep rallies going.
→Use GPU-accelerated training to iterate on the agent's assistive behaviour.
// outcomes
●An assistive partner that plays the game with the patient rather than against them.
●Adapts play to keep the patient engaged and moving through therapy.