“Powerful enough to fracture a human skull.” Those words, attributed to Robert Gruendel, the former head of product safety at Figure AI, cut through the usual optimism surrounding humanoid robotics like a warning siren. Gruendel’s lawsuit against the California-based firm alleges its machines, designed for both industrial and potentially domestic environs, can apply forces far beyond safe human thresholds, citing an incident where a malfunctioning robot’s strike left a ¼‑inch gash in a stainless steel refrigerator door.
The robot in contention is Figure’s F.02 humanoid-a second‑generation model that has already been deployed in real‑world manufacturing. According to the impact testing Gruendel conducted, the robot moved at “super-human speed” and administered blows measured at twenty times higher than the threshold of pain, generating more than twice the force required to fracture an adult human skull. In terms of industrial robotics, this puts the F.02 well outside the limits outlined in ISO/TS 15066 for safe collaborative operation, where “Power and Force Limiting” modes are intended to keep forces exerted upon contact within injury‑free ranges.
The complaint describes how warnings about safety were ignored, safety roadmaps were “gutted” before investor presentations and even hardware safeguards were removed for aesthetic reasons. “This case involves important and emerging issues, and may be among the first whistleblower cases related to the safety of humanoid robots.” said Gruendel’s attorney, Robert Ottinger. Figure, which was valued at $39 billion after a recent funding round led by Parkway Venture Capital, Nvidia and Jeff Bezos, denies the allegations, claiming it fired Gruendel because of his poor performance.
From an engineering perspective, the charges put a very strong focus on actuation systems and force‑limiting architectures in humanoid design. This is usually provided by means of impedance control, where the output of the actuator is modified with respect to the measured displacement and environmental resistance in order to achieve safe interaction. However, true back‑drivability and low inertia are difficult to achieve with high‑payload humanoids. The intrinsic dynamics of large actuators, together with fast motion profiles, may produce hazardous kinetic energy in malfunction events, as also witnessed by the statistics of industrial accidents where 88% of injuries stem from crushing or collision events.
Historical precedent underlines the stakes. So far, industrial robots have caused fatal accidents, from the 1981 case in Japan to the 2015 Volkswagen plant death in Germany. More recently, a Tesla technician alleged that a Fanuc robotic arm struck him with the force of “an approximately 8,000‑pound counterbalance weight,” rendering him unconscious. In Korea, there were 369 operator‑injured robot accidents between 2009 and 2019, with hands, head, and upper torso among the most vulnerable areas. Many incidents occurred during abnormal operations or maintenance–contexts where safety interlocks and monitored stop functions are important.
Safety certification schemes, such as the EU Machinery Directive 2006/42/EC and ISO 12100, require risk assessments related to mechanical, electrical, and ergonomic hazards. For collaborative robots, ISO/TS 15066 defines the following operation modes that avoid hazardous contact: “Safety‑Rated Monitored Stop” and “Speed and Separation Monitoring.” For a humanoid robot in practice, these modes must employ sophisticated sensor suites, such as force/torque sensors, vision systems, and proximity detectors that feed control logic capable of instantaneous torque cutoff or modification of trajectory. If such systems are removed, compliance will be compromised and the risk of injury increased, as Gruendel contends happened with F.02.
As humanoids move from controlled lab settings to the dynamic industrial floor, engineering becomes progressively more challenging. Agile manufacturing requires flexible layouts and multi‑mode operation, yet every change has the potential to require re‑certification and hazard re‑evaluation. A few emerging approaches employ safety‑related finite state machines to allow dynamic switching among collaborative modes, where the engagement of force‑limiting and separation monitoring is guaranteed in states where humans can be reached. Without them, the high-energy actuation of the humanoid turns innocuous tasks into lethal encounters.
Gruendel’s case crystallizes a broader industry tension-the race to market versus the rigor of safety engineering. As billions are poured into humanoid robotics, the pressure to demonstrate capability can tempt firms to sideline protective measures. Yet the physics remain unforgiving: kinetic energy scales with the square of velocity, and even a modest increase in actuator speed can push impact forces into lethal territory. For both engineers and investors, the result of this legal battle may redefine how the sector balances innovation with the non‑negotiable imperative of human safety.
