Sateesh Kumar

About

Sateesh Kumar is a researcher advancing the frontiers of inverse reinforcement learning (IRL) and robotics, with a focus on enabling machines to learn complex behaviors from diverse, unstructured video data. His most-cited work, "Graph Inverse Reinforcement Learning from Diverse Videos" (2022, 3 citations), tackles a critical bottleneck in robotic skill acquisition: the reliance on manually engineered reward functions. Kumar proposes a novel graph-based IRL framework that extracts reward structures from a wide variety of third-person demonstration videos, moving beyond the constrained domains typical of prior approaches. This contribution is particularly impactful for scaling robot learning to real-world environments, where diverse, unlabeled video data is abundant. By leveraging graph representations, his method captures relational and temporal dependencies in demonstrations, allowing agents to infer robust reward functions without explicit human guidance. Kumar’s work sits at the intersection of imitation learning, computer vision, and autonomous systems, offering a pathway toward more generalizable and data-efficient robot training. While his citation count is still growing, the conceptual novelty of his approach—bridging graph theory with IRL—marks him as an emerging voice in the field, with potential to influence future work in lifelong learning and cross-domain transfer for embodied AI.

Research Focus

Key Achievements

Top Papers

1. 1
   Graph Inverse Reinforcement Learning from Diverse Videos

   3 citations · 2022

Key Collaborators

Contact & Links