Research
A panel of AI researchers convened by AIhub’s coffee corner discussion series examined whether humanoid robots represent the future of embodied AI, pointing to a recent inflection in capability driven by the convergence of improved hardware, better algorithms, and new training methods. The conversation, timed to follow RoboCup 2026, offers the robotics community a snapshot of how practitioners view the shift from classical control toward learned locomotion.
Joydeep Biswas of the University of Texas at Austin described the change as a snowballing trend that accelerated over the past couple of years, producing what he called a huge inflection in capabilities. He emphasized that no single technical innovation explains the shift; rather, it required the confluence of much better hardware, significant algorithmic advances, and progress in how robots are trained and learned. The panel also included Sanmay Das (Virginia Tech), Rina Dechter (University of California, Irvine), Ioannis Erripis (AIhub), Sabine Hauert (University of Bristol), Ella Scallan (AIhub), and Marija Slavkovik (University of Bergen).
He emphasized that no single technical innovation explains the shift; rather, it required the confluence of much better hardware, significant algorithmic advances, and progress in how robots are trained and learned.
Among the concrete observations from the discussion, Biswas noted that new humanoid robots appearing at RoboCup fell down less often, and that teams were suddenly able to deploy robots without humans following them around for support. On the control side, he contrasted the field’s history with its present: humanoid robots once required classical control, whereas most modern humanoid robots now rely on learned controllers for walking. He identified the rise of sim-to-real approaches — training models in simulation and transferring them to the real world — as part of this transition.
The discussion was framed around how the technology has evolved recently and what can be expected in the years to come, though the published excerpt does not detail specific predictions or next steps from the panel. The exchange captures a moment in which learned controllers and simulation-trained policies are becoming the default for legged humanoid locomotion, replacing classical control pipelines.
Source: Robohub