Today's Feature

Figure 3’s photo op at the White House this week was a reminder that publicity still outpaces practicality in humanoid robotics. Public demos—from a Unitree humanoid chatting with children on a New York sidewalk to another clearing snow—grab headlines and likes, but they do not answer the hard questions of city life: who pays for charging infrastructure, who repairs a stalled machine at 2 a.m., and who is liable when a bot panics in a crowded street [1][2][3]. Encounters on sidewalks are social events, not engineering benchmarks. When they go well they become viral proof points; when they go wrong they provoke instant backlash. Recent field incidents—in Macau, a remote‑operated robot frightened an elderly woman and police intervened; in San Jose, a restaurant demo toppled dishes and proved hard to power down—show how fragile public trust is once robots leave the lab. Municipal regulators respond to voters’ reactions, not white papers; companies that treat street trials as marketing risk triggering swift local bans that make later deployments costly or impossible. Safety and standards are the bottleneck. Engineers debate actuators and model sizes; cities, insurers and courts want predictable failure modes, certification and audit trails. The first ISO standard for humanoids (ISO 25785) remains unfinished even as companies push into public space, according to industry briefings. That gap forces firms into costly one‑off compliance cases and hands regulators the initiative. Building enforceable tests, logging requirements and update provenance is tedious; it is, however, what turns novelty into infrastructure. Money is chasing the wrong visible metric. Investors are pouring capital into software, simulation and defense autonomy—Shield AI’s recent $2bn raise at a $12.7bn valuation is one signal of that trend—while bets on generic, general‑purpose bodies remain speculative [7]. The commercial margins are migrating up the stack: fleet management, over‑the‑air updates, trained operators, liability insurance and the services that keep machines running will capture recurring revenue long after a hardware maker’s launch coverage fades. That is why a strategic partnership announced this week to accelerate RF transceiver development—aimed at improving sensor and connectivity reliability—matters more than another demo; rugged, predictable sensing is the substrate cities will pay for (GSME/Tarana Wireless announcement, March 27). Practical engineering choices follow. Narrow, repeatable tasks let teams iterate safety cases and maintenance without endangering the public. Pilots should be civic projects: invite transit agencies, disability advocates, unions and city attorneys into trials so rules and budgets for charging stations, depots and trained operators are designed before machines are rolled out. Design for legibility—predictable speed, clear signals when confused, an accessible pause or shutdown that bystanders can understand—is as much a product decision as it is a controls problem. For students and recruiters the hire that matters is not just a brilliant control‑theorist but someone who can translate a safety case into a binder a municipal inspector understands. Data‑driven improvement is real—some companies aim to collect millions of hours of real‑world training data to close sim‑to‑real gaps—but that requires choreography with procurement, standards and insurers. Until dozens of organisations agree on what “safe enough” looks like for narrow, repeatable services and cities budget for the long tail of maintenance, glossy demos will win headlines but rarely earn permits. The next fee‑earning milestone for this sector will be a five‑year municipal contract to keep a neighborhood clear, on time and insurable—not a robot at the White House. If the industry wants robots in public life it must stop pitching miracles and start building the civic scaffolding that makes them useful.