Marc Raibert, the founder and chairman of Boston Dynamics, has introduced a remarkable assortment of both bipedal and quadrupedal robots that excel in performing astonishing parkour feats, engaging dance sequences, and adeptly stacking shelves.
Raibert is now setting his sights on spearheading a transformation in robotic intelligence as well as their acrobatic skills. He noted that recent strides in machine learning have significantly improved his robots’ capacity to master challenging maneuvers without human intervention. “We aim to generate a wide array of behaviors without needing to meticulously design every action that robots execute,” Raibert shared with me recently.
While Boston Dynamics has been a frontrunner in the field of legged robots, it now finds itself amid a bustling landscape of companies developing robot dogs and humanoid machines. Just this week, a startup named Figure unveiled a new humanoid called Helix, which reportedly can help unload groceries. Another firm, x1, showcased a strong-looking humanoid called NEO Gamma performing household chores. Additionally, Apptronik announced its intentions to ramp up production of its humanoid, dubbed Apollo. However, demonstrations can often be deceptive. Moreover, many companies do not disclose their humanoids’ pricing, leaving it uncertain how many actually envision selling them as domestic assistants.
The true measure of these robots will be their ability to operate independently from human programming and direct oversight. This capacity will hinge on advancements similar to those Raibert is advocating. Last November, I reported on initiatives aimed at developing entirely novel control models for robots. If this research begins to yield results, we may witness rapid progress among humanoids and quadrupeds.
Boston Dynamics markets a four-legged robot named Spot, which is utilized in oil rigs, construction sites, and other environments where wheeled vehicles face challenges with the terrain. The company also produces Atlas, a humanoid robot primarily for research purposes. Raibert indicated that Boston Dynamics employed a machine learning approach known as reinforcement learning to enhance Spot’s running capabilities, allowing it to move three times faster. This same technique is also enabling Atlas to walk with greater confidence, Raibert adds.
