Robotics & Embodied AI

Current landscape

NHTSA regulates autonomous vehicles through voluntary exemption and standard-amendment tools, not a comprehensive AV statute. In June 2026, it withdrew its own proposed "AV STEP" voluntary safety-disclosure program — industry groups argued the reporting burden outweighed the benefit — and the same day proposed amending federal vehicle safety standards to let vehicles with no pedals or steering wheel meet braking, shifting, and defrost requirements without human-driver hardware. Case-by-case exemption requests, like Zoox's pending petition for a two-year, 2,500-vehicle exemption from eight federal safety standards, remain the primary mechanism for novel vehicle designs. Congress has passed no federal AV statute: a bill giving NHTSA expanded design-standard authority cleared a House committee on a narrow, largely party-line vote in 2026 but hasn't passed either chamber; a separate surface-transportation reauthorization bill advanced with much broader support and includes a federal framework specifically for autonomous commercial vehicles. The result is that state law still governs most AV liability and operating questions, with roughly two dozen states having enacted some form of AV legislation and dozens more bills introduced annually.

Incidents, and what they show about the current default

2025–2026 produced an incident record. Waymo issued three separate voluntary software recalls — for not reliably stopping at school buses, for one vehicle swept into a flooded creek (triggering a fleet-wide recall of nearly 3,800 vehicles), and for vehicles entering active highway construction zones (nearly 4,000 vehicles) — all self-reported, none carrying a fine. Tesla's Full Self-Driving software is under a formal NHTSA Engineering Analysis covering 3.2 million vehicles for camera-degradation failures linked to nine incidents including one fatality, plus a separate probe covering 2.88 million vehicles for red-light and wrong-lane violations; no recall has been ordered yet. Most significantly, a Miami jury in August 2025 found Tesla partly liable in a fatal 2019 Autopilot-involved crash, awarding $329 million (including $200 million in punitive damages) and apportioning 33% of fault to Tesla and 67% to the driver — the first jury verdict holding Tesla liable for an Autopilot crash, now under appeal. That verdict is the current de facto liability framework: ordinary product-liability and negligence law, apportioned case by case by a jury, with no statutory standard guiding the split. GM's exit from the robotaxi business in 2024 (folding Cruise into an in-house driver-assist unit) was a business decision following a prior regulatory suspension, not a new regulatory action in this window.

Workplace robotics: commercial deployment, no dedicated standard

Commercial humanoid and collaborative robot deployment is underway, if still early-stage. Agility Robotics' Digit is doing paid warehouse work across nine facilities for Amazon, GXO, Toyota, and Schaeffler, logging tens of thousands of operating hours. Figure AI's humanoid robot completed a multi-month pilot at a BMW plant covering tens of thousands of vehicles before a successor model took over logistics work, with a second BMW pilot beginning in 2026. Tesla's Optimus program's deployment claims are considerably more contested — some reporting describes over a thousand units "working" in Tesla factories, while Elon Musk himself stated in January 2026 that Optimus was still "in the R&D phase" and "not in usage in our factories in a material way." That direct contradiction between promotional coverage and the company's own CEO is exactly why Proposal 3's push for standardized, audited reporting matters. Self-reported deployment claims currently can't be taken at face value even from the companies making them.

OSHA has no robotics-specific safety standard at all; its own guidance points employers to general industry rules (machine guarding, lockout/tagout) plus voluntary consensus standards (ANSI/RIA R15.06, ISO 10218, ISO/TS 15066) that carry no OSHA enforcement mechanism. A new international standard specifically for humanoid robots is in draft as of 2025 but not finalized or adopted by OSHA. Earlier research on Amazon's robotic fulfillment centers found higher serious-injury rates than non-robotic facilities; more recent analysis suggests a more mixed picture — fewer severe injuries, more minor ones — tied to pick-rate pressure rather than the robots themselves. Either way, the exposure is growing while the applicable standard remains entirely voluntary.

Trade-offs and opposing views

A federal AV framework risks federal preemption of state rules that are, in some cases, more protective than what a national floor would set — the same federal-floor-versus-state- innovation tension already named in AI-02, and resolved the same way here: a floor, not a ceiling, preserving room for states to add on. Industry's argument against the withdrawn AV STEP program — that safety-data reporting requirements impose compliance cost — is a legitimate concern to weigh, not one that should determine whether transparency happens at all; a liability framework and a safety standard are both unworkable without operational data, so the cost argument bears on how the reporting requirement is designed, not on whether one exists. On OSHA rulemaking specifically, some in industry prefer voluntary consensus standards precisely because they move faster than federal rulemaking — a fair point given how slowly agency rulemaking generally moves, but one that trades speed for enforceability in an area where the failure mode is physical injury, not a compliance memo.