Suhan Shetty

Idiap Research Institute, Robert Bosch (China), École Polytechnique Fédérale de Lausanne

Papers

Total Citations

H-Index

About

Suhan Shetty is a robotics researcher whose work lies at the intersection of ergodic control, tensor methods, and reinforcement learning for complex manipulation and locomotion tasks. His most impactful contribution is the development of tensor train-based approaches for ergodic exploration, particularly applied to insertion tasks (21 citations), where he reformulated the spectral multiscale coverage problem to enable robots to efficiently explore probability distributions over areas rather than tracking fixed trajectories. Shetty has also pioneered the use of tensor methods for global optimization in robotics (7 citations), addressing the critical challenge of initial guess dependency in numerical solvers by initializing them near global optima without requiring gradient access. In locomotion, he explored trajectory-based deep policy search for quadrupedal walking (8 citations), proposing an alternative to step-by-step policy optimization by determining optimal policies for entire walking cycles. His more recent work on robust execution of assembly policies (2 citations) introduces pose-invariant task representations to handle uncertainties in initial object poses and grasp errors, ensuring successful contact-rich assembly operations. Shetty’s research demonstrates a consistent focus on making robots more adaptive and reliable in real-world, contact-intensive environments.