Rishabh Agarwal
Papers
3
Total Citations
13
H-Index
3
About
Rishabh Agarwal is a leading researcher at the intersection of deep reinforcement learning (RL) and robotics, whose work has fundamentally reshaped how we train and scale value-based algorithms. His most influential contribution, the DR3 paper (2021, 6 citations), introduced a critical insight: that deep RL’s notorious instability stems from a lack of explicit regularization, not just overparameterization. By proposing a simple regularization penalty, DR3 provided a principled solution to a long-standing problem, offering a path toward more reliable and sample-efficient RL agents. Agarwal’s impact extends to practical robotics. His 2019 work on low-cost tactile sensing systems (4 citations) demonstrated how to integrate 3-axis force sensors into robotic grippers, enabling precise measurement of shear forces during manipulation—a key step toward dexterous, real-world grasping. Most recently, his 2024 paper “Stop Regressing” (3 citations) challenges a core tenet of RL by proposing to train value functions via classification rather than regression. This paradigm shift promises to dramatically improve scalability and stability for large neural networks, potentially unlocking new frontiers in complex, high-dimensional decision-making. Through these innovations, Agarwal is not only advancing theoretical foundations but also building the practical tools for the next generation of intelligent, embodied agents.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3