# Practical Robot Learning from Demonstrations using Deep End-to-End Training

## TL;DR

This paper presents a practical end-to-end deep learning approach for robot learning from demonstrations, enabling quick training and deployment of behaviors directly from camera images and human demonstrations.

## Contribution

It introduces a deep ResNet-based model that learns robot behaviors from visual demonstrations in under an hour, demonstrating rapid training and deployment.

## Key findings

- Learning a servoing task in less than an hour
- Only 16 minutes of demonstrations needed for effective learning
- End-to-end training from camera images is feasible and efficient

## Abstract

Robots need to learn behaviors in intuitive and practical ways for widespread deployment in human environments. To learn a robot behavior end-to-end, we train a variant of the ResNet that maps eye-in-hand camera images to end-effector velocities. In our setup, a human teacher demonstrates the task via joystick. We show that a simple servoing task can be learned in less than an hour including data collection, model training and deployment time. Moreover, 16 minutes of demonstrations were enough for the robot to learn the task.