UC Berkeley's Ctrl-F adds force commands so humanoid robots can move 20 kg loads
ChongZzZhang · x · 2026-11-06
UC Berkeley researchers present Ctrl-F, treating explicit 3-D force commands as the missing interface for humanoid loco-manipulation: reference motions say where the body should be, not how hard to act on the world.
The method has two stages:
- A low-level tracker trained without any real object: each episode samples a 3-D force command and applies the opposite wrench to a force body (hand or foot), so the only way to hold the reference pose is to produce exactly the commanded force.
- A high-level planner reads object state and proprioception to choose force commands online, adapting when the object's mass changes mid-task.
On a Unitree G1, the system drags, pushes, raises and kicks loads up to 20 kg where a force-blind tracker moves nothing. Paper and video are out; code coming soon.
More from Embodied
- Swiss startup EmbodiedAI unveils AIR, a humanoid robot for safe industrial automation — lukas_m_ziegler · 2026-11-06
- TestingCatalog starts hands-on with Surface Laptop Ultra packing NVIDIA RTX Spark N1X chip — testingcatalog · 2026-11-06
- EPFL spin-off Embodied AI unveils AIR, a legless dual-arm wheeled humanoid built for factory floors — lukas_m_ziegler · 2026-11-06
- Machenta brings a robot to London's King's Cross to let the public teach it on the spot — Ed__Johns · 2026-11-06
- Archon Robotics debuts Anchor 1, a humanoid whole-body controller cutting tracking error 30-80% — ChongZzZhang · 2026-11-06
- Lightwheel and Synopsys bring CAE-grade contact physics into robot simulation — lukas_m_ziegler · 2026-11-06