Paresh Dhakan

Papers

Total Citations

H-Index

About

Paresh Dhakan’s research lies at the intersection of reinforcement learning, robotics, and autonomous skill acquisition, with a focus on enabling agents to learn continuously and adaptively without human intervention. His major contributions include proposing general intrinsic reward functions for maintenance, approach, avoidance, and achievement goal types—a framework that reduces the need for task-specific reward engineering and allows agents to internalize diverse objectives. This work, his most cited with 9 citations, provides a foundation for more autonomous learning systems. Dhakan also advanced open-ended learning through a domain-independent goal generation mechanism, enabling robots to autonomously set and pursue increasingly complex compound goals. His research on concurrent skill composition demonstrates how ensembles of primitive skills can be combined to solve novel tasks efficiently, a key step toward cumulative learning in robotics. By tackling the challenge of learning without predefined curricula or external rewards, Dhakan’s work contributes to the vision of truly autonomous, self-improving robotic agents that can operate in unstructured environments.