Humanoids

ROBOTIS AI Sapiens Runs the Hangang River Path in 38°C Heat

Shar Hendrix 4 min read

ROBOTIS just put its open AI Sapiens humanoid on a sweaty public path. In a new OpenSourceTeam clip titled Open Humanoid ROBOTIS AI Sapiens #5, the biped runs along Seoul’s Hangang River while the caption and third-party recaps put ambient conditions around 38°C. That is a different test than a lab carpet: sun, pavement, heat soak, and continuous locomotion policy work.

AI Sapiens outdoor run along the Hangang. Source: ROBOTIS OpenSourceTeam / YouTube.

What the platform is

AI Sapiens is ROBOTIS’s open-source humanoid for physical AI, built on in-house Dynamixel-Q quasi-direct-drive actuators. Company product copy frames it as hardware meant for imitation learning and reinforcement learning policies on real metal, not only in sim.

Earlier Mechafeed coverage and ROBOTIS docs put the research body (K0/K1 class) roughly in the ~1.3 m / mid-30 kg / 23-DoF band with Dynamixel-Q legs and arms. The Hangang video is not a new SKU announcement; it is a field stress demo of locomotion policy outdoor.

ROBOTIS AI Sapiens product still
AI Sapiens product still from the ROBOTIS product site. Source: ROBOTIS.

Why the riverside clip matters

Open humanoid platforms often ship with Isaac Sim recipes and indoor gait loops. A continuous outdoor run on a real path stresses:

  • Thermal and continuous duty on QDD actuators under high ambient heat
  • Terrain and balance without a flat motion-capture floor
  • Policy transfer from training stacks (ROBOTIS has shown RL locomotion and, earlier in 2026, text-to-motion via NVIDIA Kimodo on Discourse)

Humanoids Daily and other recaps of the same clip emphasize the heat callout and the open-source pitch (BOM, CAD, source, sim assets as the program matures). Treat social recaps as leads; the primary receipt is the official YouTube upload plus ROBOTIS product/docs pages.

A Human’s Take

I’m so here for open hardware that leaves the lab before the press deck is finished. A Hangang run does not prove warehouse throughput, but it is a cleaner honesty check than another perfect indoor circle. If ROBOTIS keeps shipping CAD, firmware, and RL configs with demos like this, small labs actually get a path. What I want next: distance, fall rate, and whether the same policy still works when the river path is wet.

Sources