# Humanoids Learn Self-Other Distinction
_Humanoid robots now learn self-other distinction and build predictive self-models from sensory data, enabling better collaboration and task performance in human-robot environments._
**Published:** 2026-06-13
**Source:** https://www.startuphub.ai/ai-news/ai-research/2026/humanoids-learn-self-other-distinction
---
Humanoid [robots](/ai-news/artificial-intelligence/2026/grace-brown-on-building-robots-with-personality) increasingly operate alongside humans, yet a critical gap remains: their inability to distinguish themselves from others. This lack of self-other distinction hinders effective collaboration and safe navigation in shared spaces.

Robots lack self-other distinctionDriver

From the article 2 mentionsThis lack of self-other distinction hinders effective collaboration and safe navigation in shared spaces.problemProprioceptive-visual correspondenceCorerobot learns to differentiate itself from others using sensory dataFrom the articleResearchers have demonstrated a novel approach where a humanoid robot learns to differentiate itself from others solely through proprioceptive-visual correspondence.Bypasses identity labelsContextFrom the articleThis breakthrough bypasses the need for explicit identity labels or complex kinematic models, a significant hurdle in current robotics.Predictive self-modelCoreFrom the article 3 mentionsThe system establishes a predictive self-model that maps joint configurations to its three-dimensional body occupancy, effectively learning how its own body shape changes with movement.enablesLearned self-model instrumentalEffectenables downstream tasks in multi-agent scenariosFrom the articleOnce this foundational self-other distinction is established, the learned self-model proves instrumental in various downstream tasks.Better collaborationOutcomeimproved task performance in human-robot environmentsFrom the articleThis lack of self-other distinction hinders effective collaboration and safe navigation in shared spaces.Robust multi-agent interactionContextfundamental for effective human-robot collaboration

## Bootstrapping Self-Representation from Sensory Data

Researchers have demonstrated a novel approach where a humanoid robot learns to differentiate itself from others solely through proprioceptive-visual correspondence. This breakthrough bypasses the need for explicit identity labels or complex kinematic models, a significant hurdle in current [robotics](/ai-news/ai-research/2026/google-deepmind-fuels-european-robotics-startups). The system establishes a predictive self-model that maps joint configurations to its three-dimensional body occupancy, effectively learning how its own body shape changes with movement.

## Enabling Robust Multi-Agent Interaction

Once this foundational self-other distinction is established, the learned self-model proves instrumental in various downstream tasks. In scenarios involving multiple agents, including humans and morphologically identical robots, the system reliably identifies itself. This capability directly supports critical functions such as target reaching, collision-aware motion planning, and human-to-robot motion retargeting. The ability to form a 3D self-model is a crucial step towards true humanoid robot self-awareness.
---
Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.