1. Safety in robotic systems
My interests in real-time safety critical control gradually increased as I worked more and more with industrial robot arms. As we see an increasing use of robots and robotic technology around us, safety will be of critical importance. We need guaranteed margins of safety in order to create a human-friendly environment. My focus currently has been to introduce robustness in safety-critical control via the notion of input-to-state safety. This was mainly motivated from my PhD work on input-to-state stability. Follow the link below for more details:
S. Kolathaya and A. D. Ames. Input-to-State Safety With Control Barrier Functions. 2018. IEEE Control Systems Letters. Volume 3, Issue 1, Pages 108-113. .pdf .bib
2. Legged locomotion
RBCCPS has a custom built four legged robot (quadruped) called “Stoch”. The main goal is to explore deep-reinforcement learning algorithms to achieve a wide variety of walking behaviours. The quadruped must be able to plan and navigate over a diverse set of environments like flat plains, undulating and slippery surfaces, and also muddy terrains. See video below of our most recent walking result! Gaits were generated via D-RL based training, and then implemented via kinematic motion primitives.
More details are in this paper:
Abhik Singla, Shounak Bhattacharya, Dhaivat Dholakiya, Shalabh Bhatnagar, Ashitava Ghosal, Bharadwaj Amrutur and S. Kolathaya. Realizing Learned Quadruped Locomotion Behaviors through Kinematic Motion Primitives. 2019. in IEEE International Conference on Robotics and Automation. .pdf .video