Research Projects

The proposed project aims to develop Iterative Learning from Demonstration (ILfD) method to enable robot-based electric vehicle (EV) battery disassembly tasks. Several countries, including India, are pursuing electrification of transportation systems through EVs to achieve climate protection and emission reduction goals. EV battery pack forms an important component contributing to over 50% of EV value, and will also form a large part of EV waste in the coming decades. Furthermore, end of life (EoL) EV battery consists of several components which can be reused or put to secondary uses such as storage systems promoting sustainability. Efficient non-destructive disassembly is required for achieving the recycling of EV batteries at scale. Robot-based disassembly of EV batteries can be a potential solution as it provides efficiency at scale. It also provides a safer option to manual disassembly for interacting with high voltage and potentially flammable battery components. However, the control of robots for EV battery disassembly is challenging because of the complexity of the tasks and variability of battery configurations and layout. The proposed project aims to develop robot learning approaches for complex disassembly tasks such as unscrewing and manipulation of deformable components such cables and plugs, with ability to generalize over different battery configurations. Learning from Demonstration (LfD) is a promising approach through which robots can be taught complex disassembly operations from human demonstrations, without manual programming which is infeasible for complex operations. Teleoperation can be used to demonstrate the disassembly task sequences remotely by human operators with force, torque and vision feedback. The demonstrated force, torque and motion data can be used to learn trajectories for implementing the disassembly tasks. However, teleoperation tends to be imperfect due to noise and delay in force and vision feedback, and reduction in demonstration precision due to remote operation. In order to overcome the limitation of imperfect demonstrations, iterative learning approaches can be used to improve the LfD for robotic disassembly. Therefore, this project proposes to develop Iterative Learning from Demonstration (ILfD) to improve skill transfer from human operators to robot through iterative refinement of task models. An uncertainty mitigation approach will be developed, where the errors in robot learning, represented through estimated uncertainty along task trajectories, can be iteratively reduced through control and task model updates over successive demonstrations. The proposed approach will be experimentally validated and accurate simulation models will be developed for unscrewing and deformable object handling operations, involving a robot arm with disassembly tool and a human operator with a teleoperation device consisting of force and vision feedback sensors.