E-TTS: Embodied Test-Time Scaling for robotic manipulation

arXiv·medium signal

This arXiv paper (2026-06-25) proposes E-TTS, a test-time scaling framework for embodied robotic manipulation that studies how reasoning-based scaling improves policy performance and incorporates historical context for long-horizon, sequential tasks rather than relying only on current observations. It addresses two open gaps: an under-studied scaling mechanism for embodied reasoning and the need for history utilization in action scaling. The work extends the test-time compute paradigm from text into robotics, relevant to builders working on embodied agents.