MIT's Russ Tedrake Says Robotics Is Finally on a Rocket Ship

When Russ Tedrake saw his little robot learn to walk on its own in 20 minutes, it wasn't just a lab experiment: it was a sign that robotics was about to shift gears, and he himself admits, "the success of machine learning has gone much faster than our ability to really understand it." Until a few years ago, building a robot that could walk was a technical challenge for old-school engineers: everything was designed according to precise principles, perfect control was sought, and every misstep was a step backwards. Today, however, Tedrake says that the role of engineers is changing: we are no longer just builders, but we are becoming behavioral scientists who observe what happens when the system learns on its own — and often we don't really know why it works so well. The argument is this: after decades of promises, robotics is now truly on the launchpad. Not because we have finally understood all the secrets of the physics of walking or of artificial intelligence, but because the way machines are built and learn has changed radically. We no longer teach robots every detail: we start from general models, such as those that already dominate language or video, and we "train" them only to take the extra step, to connect general common sense with physical action. Tedrake has deep roots: the son of a General Motors engineer in Detroit, he spent his adolescence learning the basics of automation in a Ford factory, where a coding error he made—stopping the fans when a cable was disconnected—caused the temperature to rise above the union limit and shut down the production line. "They yelled at me, but that time I really understood what it means to shut down a factory," he says. Later, at university, robotics was scarce and he found his way by working on the artificial intelligence of video games, before falling in love with two-legged robots in the MIT labs. His thesis: the most efficient walking robots were not those full of motors and controls, but those that exploited physics, like certain toys that go down a ramp with a slight push, letting gravity do most of the work. And here is the first fact that turns everything upside down: reinforcement learning models, seen for years as little more than a nice idea, have become the engine of new robotics in just a few years, simply because the available computing power and the ability to simulate millions of scenarios have made it possible to train robots almost like you do with a video game. Tedrake puts it bluntly: "We ended up with systems that worked incredibly well before we even really understood why." This leap has transformed the profession: it is no longer a matter of designing everything down to the smallest detail, but of observing, experimenting, seeing what happens when the robot tries on its own — and only then trying to understand the rules that emerge. One of the most human moments comes when Tedrake describes how robotics has become an open discipline: he says that today anyone with talent can get there, no matter if they come from the automotive, medical, or software fields. And Boston, with its ecosystem of startups and laboratories, is the place where this cross-pollination is most evident. Another reversal: many think that the problem with robotics is the lack of data — that it cannot compete with the mass of information on which large language models have been trained. Tedrake, on the other hand, explains that the real breakthrough is building a bridge: starting from models that are already full of common sense about the world and "teaching" only how to translate that knowledge into specific physical actions. There is no need to rebuild everything from scratch. This makes robotics much more scalable and less dependent on huge proprietary datasets. And when it comes to the future, Tedrake is clear: "There's no guarantee that we'll make it this time, but I'd rather be on the spaceship heading for the moon than stay on the ground." But there is a note of caution: if technology changes the very nature of work, there is a risk that the sense of personal value that so many people find in their profession will be lost. For this reason, he says, we need to approach everything with empathy, listening to those who will be involved and learning from those who have already seen similar revolutions — such as in software or graphics — to prevent the transformation from leaving behind those who currently work with their hands. Here's the final thought: the real revolution in robotics is not making smarter machines, but rethinking the way we learn from the machines themselves. If you find yourself looking at technology with different eyes after this story, you can mark that change on Lara Notes with I'm In: it's not approval, it's a declaration that this perspective now belongs to you. And if you happen to tell someone how robotics is changing both work and humanity, you can tag that conversation with Shared Offline: because some ideas need to be captured when they become real conversations, not just links. This Note comes from the Automated podcast and saves you 43 minutes of listening.