Meta Partnr: Robots to take over tedious tasks in future

Robots will soon be able to act socially in a simulation and then also in the real world. At least, if Meta has its way. They have already tried this out with Spot, the dog-like robot from Boston Dynamics. Spot acts autonomously and is social at the same time. Robots should become partners, not just agents.

At the celebration of the tenth anniversary of the FAIR team at Meta (Fundamental AI Research), Spot walks through the hiply designed rooms of the Meta office in central Paris. In blue, unlike the yellow original, Spot picks things up, tidies up and is cute – Putting all thoughts of creepy, autonomous robots with evil missions to one side.

Last but not least, Spot is given a master or mistress with a quest, Meta's VR headset. This allows you to look into the robot's head and see what it sees and read what it wants to do. Its task in the FAIR office: to move toys from A to B. They are scattered everywhere. Plush pineapples, plush donuts, board games. Spot has to go in search of the pineapple. He remembers where it was. But the nasty mistress, a Meta employee, puts it somewhere else just before Spot finds it. He has to search again, finds it, grabs it and follows the task of moving it to another place.

The robot, your better partner

Meta dreams of robots helping with cooking as a matter of course in the future and also anticipating other needs. However, even the people closest to you normally find it difficult to – smell what someone wants.

Partnr is the name of the framework that Meta offers as open source. A path that the company has generally taken in the field of AI – even if not everything is always freely available. To date, Meta has published more than 1000 AI models as open source. Joelle Pineau, Vice-President AI Research, gives a whole talk at the ceremony on why Meta believes it can achieve more by making models freely available. At the end of the day, it's the usual advantages of open source. Pineau and the team have a very scientific approach to their work.

Partnr also includes a freely available data set and a benchmark with thousands of tasks. Using Partnr, robots can first be trained in a simulation and then set loose in the real world. Partnr is based on the various Habitat simulation platforms from Meta. In these, robots can be navigated in a 3D world.

Transparency note: The author was invited by Meta to the FAIR anniversary celebration in Paris. Meta covered the travel costs. There were no specifications regarding the nature and scope of our reporting.

(emw)