Today Mohak Bhardwaj, member of Air Lab, presented his paper on Learning Heuristic Search via Imitation at CORL in Mountain View, CA. His work advances the state of the art for motion planning, minimizing search effort in graph-based planning environments. The video and paper can be found at:
Sankalp Arora is a doctoral candidate Ph.D. student at the prestigious Robotics Institute (RI) at Carnegie Mellon University. He holds a Masters in Robotics from RI and undergraduate in electronics from Delhi College of Engineering. He also has experience working with Hitech Robotics Systemz. During his graduate studies at RI he has developed planning and perception algorithms for flying vehicles. He developed safety and sensor planning for AACUS, the world's first guaranteed safe full scale autonomous helicopter and the world's first weather invariant infrastructure free ship-deck tracker. He is a recipient of the 2016 Qualcomm Innovation fellowship for the development of autonomous Unmanned Aerial Vehicles for data gathering. Sankalp is now working towards developing AI for flying vehicles for high value commercial applications like inventory taking, inspection and logistics. ... See MoreSee Less
Air Lab's paper on Learning Heuristic search via Imitation has been accepted at Conference on Robotic Learning (CoRL www.robot-learning.org/), 2017! It is one of the 18 papers selected for a full length talk.
Congratulations to Air Lab members Mohak Bhardwaj, Sanjiban Choudhury and Sebastian Scherer for the acceptance!
The paper provides a novel formulation of search based planning as sequential decision making and presents the Search as Imitation Learning (SaIL) algorithm, that efficiently trains a heuristic policy for minimizing search effort, using a framework of cost-senstive imitation learning. SaIL demonstrates interesting results on different distributions of planning problems and outperforms current state of the art. ... See MoreSee Less