Project Mentor: Prof. Mayuresh V. Kothare, Chemical & Biomolecular Engineering
2027 Impact Fellows:
Project Description:
Self-driving laboratories (SDLs) are transforming chemical and materials discovery through closed-loop automation, yet automated infrastructure for physical manipulation of soft, deformable matter remains outside the reach of current robotic platforms. Our group recently developed DropletRunner, the first robotic platform for autonomous liquid droplet navigation on an open, unconfined surface, where a model-based reinforcement learning (MBRL) agent using DreamerV3 learned the task from limited physical episodes with no simulator. Learning directly on hardware in this way, with no simulation in the loop, is both technically demanding and largely unexplored for physical systems.
This project extends the platform to multi-droplet coordination, each assigned a distinct target and a distinct path geometry, and the central difficulty is that a single command acts on every droplet at once. We will formulate a multi-agent reinforcement learning (MARL) problem in which agents learn to coordinate the timing and direction of control actions to navigate the droplets to their own goal. The project involves extending the existing world model to the multi-agent setting, designing reward structures that assign credit to individual droplets, and continuing offline training. The experimental setup consists of camera-based state estimation, ROS2 asynchronous communication, motor actuation, and a learning loop running on the physical system.
Integrating MARL, mechanical actuation, and microfluidic manipulation into a single hardware-in-the-loop pipeline, this work offers exposure to a niche of system-level experimental design. The researcher will develop cross-disciplinary fluency in real-time control, sample-efficient learning on physical hardware, and robotics middleware, precisely the integrative skill set that next-generation SDLs demand.