A tutorial session on Efficient Batch Optimization for Planning and Control by Dr. Arun Kumar Singh.
Trajectory optimisation is one of the bedrocks of modern motion planning in robotics as it allows users to encode higher-level behaviours through carefully designed cost and constraint functions. In this talk, he will discuss a special premise where one needs to solve hundreds of mildly related variants of a trajectory optimisation problem in parallel: a.k.a, the batch setting. He will also discuss examples from autonomous driving, drone cinematography, and crowd navigation to show how:
- Batch optimisation can counter the inherent non-convexity of the many trajectory optimisation problems.
- It can be combined with gradient-free sampling techniques like cross-entropy methods.
- It can be used to construct a new class of bi-level optimizers.
Dr. Arun Kumar Singh is an Associate Professor at the Institute of Technology, University of Tartu, Estonia, where he leads the Collaborative Robotics and Robotic Computing Group. Before joining the University of Tartu, he has been a postdoc at NTU, Singapore; Tampere University, Finland and BGU, Israel. His research interests lie in the algorithmic foundations of robotics, especially in the context of motion planning, control and learning for control.
Date: 15 November 2022
Time: 5:00 pm
Venue: H-304, Himalaya Block, IIIT Hyderabad
Page last updated on November, 2022