Schedule
Day 1: Intro and Lectures (3rd March 10:00 – 14:00)
- [Aamir] Introduction and overview of the workshop topic
- Topic 1: [Eric] Exploiting (supervised) learning-based detectors for on-board, realtime localization and target tracking. Lecture (1 hr): IROS 2018 paper
- Topic 2: [Rahul] MPC and reinforcement learning based formation control methods.
Lecture (1 hr + questions): IROS 2019 + IEEE-RAL 2020
- Topic 3: [Yu Tang] Model-free residual reinforcement learning for blimp control.
Lecture (1 hr + questions) ICRA 2022 + IROS 2022
Two
of the above topics will be followed by a tutorial. Each tutorial will
be centered around a case-study which we will design. The tutorial
session will introduce the problem to the students, teach
how to address it, program etc.
Day 2: Tutorials (4th March 10:00 – 15:00)
-
[Eric] Tutorial (2 hr): How to deploy an existing deep neural network on a robotic platform
-
Requirements:
-
Desktop computer, Nvidia GPU or a powerful gaming laptop
-
Ubuntu 20 LTS, ROS Noetic (Python 3).
-
Cuda
-
PyTorch
-
Webcam (for realtime visual detection)
-
[Yu Tang]
Tutorial ( 2-3 hrs) Basic RL, setup of RL agent, intro to PPO how to
write a PPO agent + a robotic task example – navigate a turtlebot to a
position.
Day 3: Demo (7th March 10:00 to 14:00)
- [Eric] Demo by students (2 hr): DNN-based detectors on quadcopter hardware
-
[Yu Tang] Demo by students of the robotic example
The students should present the results of their exercises, and lessons learned during the demo