Shuijing725 / CrowdNav_DSRNN

[ICRA 2021] Decentralized Structural-RNN for Robot Crowd Navigation with Deep Reinforcement Learning
https://sites.google.com/illinois.edu/crowdnav-dsrnn/home
MIT License
99 stars 23 forks source link

run on turtlebot2 #13

Closed asialulu closed 2 years ago

asialulu commented 2 years ago

hi, I'm very interested in experiments on robots. What do I need to do to run on robots? Thank you!

Shuijing725 commented 2 years ago

Hi asialulu,

There are several pieces that are needed to fill the sim2real gap:

  1. Use sensors to detect the positions of people;
  2. Track the ID of each detected person so that they are in a consistent order, and feed the positions of tracked people into the observation of the RL network;
  3. The robot state can be obtained by running SLAM;
  4. Use a more realistic robot dynamics, such as unicycle (set action_space.kinematics = "unicycle" in crowd_nav/configs/config.py). However, the true dynamics depend on the robot you are using.

Hope it helps!

Shuijing725 commented 1 year ago

@asialulu The sim2real tutorial is released here: https://github.com/Shuijing725/CrowdNav_Sim2Real_Turtlebot