Implementing Reinforcement Learning for Robots using Tensorflow.js

Hello, has anyone used tensorflow.js / ml5.js / danfo.js / magenta.js etc. to implement reinforcement learning applications for robots?

I have not personally however there are groups of folk interested in RL with TensorFlow.js. They have a discord server.

Also on the robotics side, Cylon.js is most popular to get at sensors and hardware which you could then use to stream to some device running TFJS to perform inference on sensor data or the like.

I have explored sensor streaming with robotics on a local network (for low latency) so the higher level “brain” if you will is on a laptop running TFJS on Chrome and then use websockets / WebRTC or similar to stream data in to make decisions sending commands back to then control the robot - offloading battery expensive compute frm the robot to the local network instead - so still ensuring privacy and even allowing offline usage as local WiFi network could be used.

Thank you, Jason, for your informative response.

Could you please provide the name of the Discord server for the group interested in RL with TensorFlow.js?

My current plan is to use TFJS for simulation in WebXR (AR) without involving hardware. Do you have any recommendations for JS libraries for RL training of robots in 3D environments? Thank you again.

Here is a discord invite to their server (3rd party not Google owned) I think the link will expire in a few days so click soon but yes those folk are interested in RL within JS: Deep Reinforcement Learning in Javascript

Thank you so much Jason !!!