Rishav Ganguly
I am a dual degree student in Mechanical Engineering at the Indian Institute of Technology Kanpur, working in Robot Learning.
I want to develop robot learning systems that can perform useful long-horizon manipulation tasks in complex environments. I am interested in combining classical robotics, reinforcement learning, imitation learning, and vision-language-action models to build efficient robot policies that can learn from demonstrations, adapt from interaction, and remain reliable on real hardware.
My current work involves low-cost robot data collection with SO101 arms, Action Chunking Transformer policies for manipulation, recurrent PPO for partially observable environments, and simulation workflows in Isaac Lab. I also write technical notes on papers, algorithms, and engineering lessons that connect mathematical ideas to working code.
email | resume | google scholar | github | blog
News
- [May 2026] Started this research portfolio and blog to document robot learning notes, paper explanations, and project build logs.
- [May 2026] Working on long-horizon manipulation experiments using ACT-style policies and SO101 teleoperation datasets.
- [May 2026] Exploring recurrent PPO for partially observable box-pushing environments with sparse rewards and local sensing.
- [May 2026] Preparing project notes on vision-language-action models, action decoders, and imitation learning for robot manipulation.