Sign In

Communications of the ACM

ACM TechNews

AI Approach Yields 'Athletically Intelligent' Robotic Dog

View as: Print Mobile App Share:
The key advance is that the robodog is autonomous, able to size up physical challenges and imagine, then execute, a broad range of agility skills based simply on the obstacles it sees before it.

A dog-like robot climbs over an unknown obstacle using a simple algorithm that encourages forward progress with minimal effort.

Credit: Shanghai Qi Zhi Institute/Stanford University

An international team led by researchers at Stanford University and China's Shanghai Qi Zhi Institute developed a vision-based algorithm that enabled low-cost, off-the-shelf quadruped robots to move from obstacle to obstacle autonomously and with agility.

The robodogs used images from a mounted depth camera and machine learning to determine whether to move over, under, or around the obstacles.

Without any real-world reference data, the robodogs learned how to navigate obstacles using reinforcement learning, with a simple reward system based on the distance they moved forward and the amount of effort exerted.

Stanford's Chelsea Finn explained, "Eventually, the robot learns more complex motor skills that allow it to get ahead."

Tests showed the robodogs could scale obstacles 1.5 times their height, jump over gaps 1.5 times their length, crawl under thresholds that were three-quarters of their height, and slip through crevices thinner than their width.

From Stanford News
View Full Article


Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account