arXiv:2606.13222v1 Announce Type: cross Abstract: Distinguishing self from others is a prerequisite for social intelligence, yet humanoid robots that increasingly share workspaces with humans still lack this ability. Here we show that a humanoid robot can learn self-other distinction from proprioceptive-visual correspondence, without any identity labels or kinematic models. Once established, this distinction bootstraps a predictive self-model that maps joint configurations to three-dimensional body occupancy, capturing how the robot's body changes with action. In multi-agent scenes involving h
Advances in AI, particularly in perception and learning, are enabling robots to achieve increasingly sophisticated social and cognitive abilities, moving beyond purely mechanical tasks.
This research outlines a fundamental step towards creating more autonomous and socially intelligent humanoid robots capable of safely and effectively operating in human environments.
Humanoid robots can now develop a rudimentary sense of 'self' without explicit programming or supervision, which is crucial for advanced human-robot interaction and collaboration.
- · Humanoid robotics developers
- · AI software companies
- · Logistics and manufacturing sectors
This enables humanoid robots to better navigate complex, multi-agent environments by distinguishing their own actions and body from others.
The development of robust self-models will lead to more adaptable and general-purpose robots, accelerating their deployment in diverse applications.
Enhanced self-awareness in robots could eventually raise ethical and philosophical questions regarding robot rights and consciousness as their capabilities mature.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI