Boston Dynamics and Ex-CEO Unite to Boost Atlas Robot's Learning

Boston Dynamics just dropped some big news on Wednesday—they're teaming up with the Robotics & AI Institute (RAI Institute) to supercharge the learning capabilities of their electric Atlas humanoid robot. This collaboration is all about diving deep into reinforcement learning, a technique that's kind of like learning through trial and error, much like how we humans and animals pick up new skills. The RAI Institute, which used to be called The Boston Dynamics AI Institute, was set up in 2022 by Marc Raibert, the same guy who founded Boston Dynamics and led it as CEO for 30 years. Raibert, a former MIT professor, is now using the Institute to keep pushing the boundaries of the research that laid the groundwork for Boston Dynamics.

Both organizations have a connection to Hyundai. The Korean car giant snapped up Boston Dynamics back in 2021, and they're also the ones funding the Institute. This setup gives Raibert the freedom to explore some pretty wild and cutting-edge tech that might not fly in a typical commercial setting. It's a bit like what Toyota did with their Toyota Research Institute (TRI), which recently teamed up with Boston Dynamics to work on large behavior models (LBMs).

These partnerships are all about making Boston Dynamics' electric Atlas better at picking up new tricks. The deal with the RAI Institute zeroes in on reinforcement learning, which can be a real time-suck but has gotten a lot more efficient thanks to simulations that let you run multiple processes at once in a virtual world.

The Boston Dynamics and RAI Institute collaboration kicked off earlier this month in Massachusetts. It's the latest in a string of joint projects, including one where they worked together on a reinforcement learning research kit for Boston Dynamics' Spot robot, you know, the one that looks like a dog. The new focus is on taking what they learn in simulations and applying it in the real world, plus figuring out how to make Atlas move and interact better in physical spaces.

When it comes to the latter, Boston Dynamics is talking about things like "dynamic running and full-body manipulation of heavy objects." These are the kinds of moves that need your legs and arms to work together. The humanoid's two-legged design brings its own set of challenges and opportunities, different from what they face with Spot. Plus, everything the robot does has to deal with a whole bunch of forces like balance, resistance, and motion.

In a statement, Raibert said, "Our aim at RAI is to develop technology that enables future generations of intelligent machines. Working on Atlas with Boston Dynamics enables us to make advances in reinforcement learning on arguably the most sophisticated humanoid robot available. This work will play a crucial role in advancing the capabilities of humanoids not only by expanding its skillset, but also streamlining the process to achieve new skills."

This news comes right after Figure AI's founder and CEO, Brett Adcock, announced that they're ditching their partnership with OpenAI to focus on building their own AI models in-house. Adcock told TechCrunch, "We found that to solve embodied AI at scale in the real world, you have to vertically integrate robot AI. We can't outsource AI for the same reason we can't outsource our hardware." Figure figured out that the best AI for their humanoid robots is the kind they develop themselves, tailored specifically for their needs. OpenAI's approach to embodied intelligence, given their size and focus on ChatGPT, just wasn't cutting it for Figure. And, just to add to the drama, there are whispers that OpenAI might be looking into making their own humanoid robots.

Most companies in the humanoid robot game are cooking up their own special AI models. That's definitely true for Boston Dynamics, with their decades of experience in crafting software for their unique robots. Even though the RAI Institute is a separate entity, they share a parent company, a founder, and, you'd think, similar goals.