The Robotics and AI (RAI) Institute has shown off a speedy version of Boston Dynamics Spot quadroped robot with advancements that have been achieved through the use of reinforcement learning.

Spot is an agile quadruped robot that comes with an API that can be used to control the way Spot moves, and the RAI has now shown off just how well this has enabled Spot to run faster. RAI used to be the Boston Dynamics AI Institute, so has close historical ties to Boston Dynamics.

In the video on YouTube, the RAI Institute says: 

"Using reinforcement learning, we trained policies for Spot that allow the robot to achieve record running speeds of 11.5 mph (5.2 m/s) — over three times faster than Spot's default max speed."

You can see the results below:

The video follows a talk given by Marc Raibert at the ICRA@40 conference in Rotterdam last November, where he explained to delegates how the new API, combined with reinforcement learning, enable Spot to be three times faster than the factory speed. 

In an article for the IEEE, Evan Ackerman said:

"If Spot running this quickly looks a little strange, that's probably because it is strange, in the sense that the way this robot dog's legs and body move as it runs is not very much like how a real dog runs at all."

Ackerman reported that Farbod Farshidian, roboticist at the RAI Institute said the gait is not biological, but then the robot isn't biological. Farshidian continued:

"Spot's actuators are different from muscles, and its kinematics are different, so a gait that's suitable for a dog to run fast isn't necessarily best for this robot."

The way the researchers managed to make Spot run faster is by moving away from the software model that comes as standard with the robot. This is based on model predictive control (MPC), so essentially is a software model attempts optimization in real time. This provides a predictable and reliable method of control, but isn't good for achieving the best possible results. The real time element is a major constraint. By moving to using Reinforcement learning (RL), the researchers could move the problem offline for the learning stage, meaning they could create as complex a model is required, and let the training take as long as was necessary to develop a control policy that could then simply run on the robot.

Farshidian explained to Ackerman that they used the model to look at what was really limiting the performance of Spot, and discovered that it was that Spot's batteries couldn't supply enough power.

Farshidian says:

"If we had beefier batteries on there, we could have run faster. And if you model that phenomena as well in our simulator, I'm sure that we can push this farther."

More Information

Reinforcement Learning Triples Spot’s Running Speed

Related Articles

Spot With AI - The New Robotics

Unitree G1 - See How It Runs

30 Years of Boston Dynamics

A Tale of Three (Robot) Dogs

Spot + ChatGTP - It's Amazing

Spot The Robot Dog Learns New Tricks

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info