Shuijing725 / CrowdNav_DSRNN

[ICRA 2021] Decentralized Structural-RNN for Robot Crowd Navigation with Deep Reinforcement Learning

Home Page:https://sites.google.com/illinois.edu/crowdnav-dsrnn/home

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Can I know how did you transfer the policy learned in the simulator to a real-world TurtleBot 2i/ 3???

sumedhreddy90 opened this issue · comments

I would be also interested

Hi all,

For sim2real, you'll need a people detector and tracker to obtain the (x, y) positions of humans. The choice of detector and tracker depends on the sensors (camera or lidar, on-board or external, etc).
In addition, we also used a more realistic robot dynamics (unicycle with noises), and tried to reduce the gap between the behaviors of simulated people and real people.

I will consider releasing sim2real code for my next paper on crowd navigation, since quite a few people are interested. But I apologize that I don't have a specific date on when this will happen.

Follow-up questions are welcome if you have any.