Autonomous robot car control demonstration in CES2016

See explanation video in https://twitter.com/Toyota/status/685657676961873920

We demonstrate how cars can learn to drive by themselves. In this demo, cars use only raw sensor data (simulated ridars: 32 distances, and angles) to control their steers and speed by deep reinforcement learning. Cars do not know their positions, and need to understand their environments only from raw sensor data and need to decide optimal control. (A QR code on the car is used for simulating ridar information). We don't give any driving rules beforehand. All these driving techniques are learned by themselves from their experiences.

A red car is manually controlled by human. Other cars try to avoid collisions.

Their learning models are shared among cars in realtime.

This is joint work with Toyota and NTT

Видео Autonomous robot car control demonstration in CES2016 канала Preferred Networks