When the Robot Bought a Plane Ticket

A 70-pound humanoid named Bebop delayed Southwest Flight 1568 out of Oakland for about an hour on April 30, 2026, because its installed lithium-ion pack busted the cabin limit. Two weeks later a smaller humanoid named Stewie walked through Harry Reid International, boarded a 737, and flew to Dallas Love Field in its own purchased window seat. Two days after that, Southwest banned humanoid and animal-like robots from the cabin and from checked baggage outright. The headlines treated it as a novelty. To anyone who has ever written a transport-mode hazard analysis, it reads as a missing artifact.

This is a robotics post, not an aviation post. The interesting failure is not at the gate. The interesting failure is upstream, in the humanoid-robot product file, where the operating-mode list almost certainly does not contain a row for "stowed in a Class C cargo compartment at 8,000 ft cabin altitude, then resting on a passenger seat at 35,000 ft." The robot was built to walk, balance, and demonstrate. It was not built — in the contractual sense of having a derived requirement set with verification evidence — to travel.

The same gap shows up in every humanoid program I have audited in the last two years. Below is the story, the standards that already cover it, a worked hazard row, a worked FMEA snippet, and five derived requirements with stable IDs.

1. The public record

April 30, 2026 — Oakland (OAK) to San Diego (SAN), Southwest Flight 1568. A four-foot, roughly 70-pound humanoid robot called Bebop traveled with its handler, who had purchased the robot its own seat. At the gate, Southwest crew flagged Bebop's installed lithium-ion battery as exceeding the 100-watt-hour cap for installed batteries in a portable electronic device under 49 CFR 175.10(a)(18). One source familiar with the gate hold described the pack as roughly four times the cabin limit. The battery was removed, the robot flew without it, and the flight pushed back about an hour late. (Fox News, eWeek, Interesting Engineering, May 2026.)

May 10–11, 2026 — Las Vegas (LAS) to Dallas Love Field (DAL). Aaron Mehdizadeh, who runs The Robot Studio in North Dallas, bought a ticket for a 3.5-foot humanoid called Stewie. Stewie was refitted with a smaller battery deliberately sized to pass TSA screening and the cabin-installed limit. The robot walked through the terminal under its own power, boarded a Boeing 737, and took a window seat. No delay, no incident, plenty of phones out. (CBS News Texas, Aviation A2Z, May 2026.)

May 13, 2026 — companywide ban. Southwest issued an internal safety alert and a policy update: no humanoid- or animal-like robots in the cabin, and none as checked baggage, regardless of size or purpose. The carrier framed it as a clarification of its lithium-battery handling policy rather than a robotics policy. (Aviation A2Z, May 15, 2026; Simple Flying, May 2026.)

Two incidents, two weeks, one ban. The reporting fixates on the optics — "the robot bought its own seat" — but the operative fact in both events is the same: a battery system whose energy budget and mode-of-carriage was decided by the robot designer, not by the regulatory framework that governs how that battery gets from city A to city B.

That is the artifact that is missing.

2. The standards lens

This story sits across four standards families. None of them, on their own, told the robot designer "stop." All four of them, taken together, should have.

2.1 The hazmat rule that actually bites

The binding rule is 49 CFR 175.10(a)(18), the DOT exception that lets passengers carry portable electronic devices containing lithium-ion cells onto a U.S. commercial passenger aircraft. The numbers are not subtle:

Installed in a device, carried in cabin or checked: lithium-ion cells must not exceed a Watt-hour rating of 100 Wh.
Installed exceeding 100 Wh but not 160 Wh: allowed only with prior approval of the operator.
Spare (uninstalled) batteries: carry-on only, individually short-circuit protected, no more than two per traveler in the 100–160 Wh band, and each spare ≤ 160 Wh.
Over 160 Wh: forbidden as passenger carry-on or checked baggage. Period. Ships only as Class 9 hazmat with a Dangerous Goods Declaration.

FAA PackSafe operationalizes the same numbers for passengers, and the IATA Dangerous Goods Regulations (Section II of Packing Instructions PI 965–970) and ICAO Technical Instructions Doc 9284 push the same thresholds out to every operator in every state that has signed the Chicago Convention. There is no jurisdiction in commercial passenger aviation where a humanoid with a 400-Wh installed pack is legal in the cabin without a Dangerous Goods Declaration, an operator approval, and a packaging spec that no humanoid product I know of has on its bill of materials.

2.2 The transport-test rule the cells already have

Every lithium cell or battery shipped commercially must carry a UN Manual of Tests and Criteria Section 38.3 test summary. The eight tests (T1 altitude simulation, T2 thermal cycling +72/−40, T3 vibration 7–200 Hz sine sweep, T4 shock, T5 external short, T6 impact / crush, T7 overcharge, T8 forced discharge) are designed to bound what the pack will see in international transport.

This is the cleanest part of the supply chain. Any reputable cell vendor will hand you the UN 38.3 test summary on request, and any humanoid-robot integrator who built a battery pack from commercial 21700 or pouch cells inherits a clean T1–T8 envelope at the cell level. The hole is one level up: a pack assembled at 400 Wh, with that specific BMS, in that specific structural mount, with that specific thermal path, did not necessarily get re-tested as an assembly, and almost certainly does not have a transport classification (UN 3480 for batteries shipped alone, UN 3481 for batteries installed in or packed with equipment) printed on the chassis.

The robot rolled up to the TSA checkpoint without the equivalent of a serial-plate.

2.3 The robot-safety standard that almost covers it

ISO 13482:2014 is the published service-robot safety standard, with an FDIS revision in approval to replace the 2014 edition. Its scope explicitly includes mobile servant robots, physical-assistant robots, and person-carrier robots — exactly the class Bebop and Stewie fall into. Clause 5.7 ("Hazards due to operational environment") and Clause 5.13 ("Hazards due to incorrect autonomous decisions and actions") both touch the right shape of problem, and Annex A's hazard list mentions transportation environments in passing.

But ISO 13482 is written for the intended use, which in practice means "operates in a home, an airport terminal, a retail floor." It does not require a transport-mode operating-state row, it does not call out hazmat conformance flowdown, and it does not require a battery-mode supervisor that can prove a "transport-safe" state has been entered before the chassis is handed to a courier. ANSI/A3 R15.06-2025 (the just-published U.S. national standard for industrial robot safety, harmonized to the 2025 edition of ISO 10218-1/-2) is even further off — it explicitly removed the term "collaborative robot" and is now organized around industrial applications. Humanoids in airport terminals are not in scope.

The standard that would be in scope, ISO 25785-1 for dynamically stable robots, is still in development. ASTM F45's mobile-manipulator performance standard is in committee. The result is what the regulators politely call a "gap": ISO 13482 is too narrow on transport, ANSI/A3 R15.06 is the wrong scope, ISO 25785-1 is not done, and 49 CFR 175 doesn't care what your product is called — only what your cells store.

2.4 The risk-management bridge

ISO 14971-style risk thinking (the medical-device discipline I keep recommending for robotics, even though it isn't formally required) closes the gap. The transport mode is an operating mode. The mode has a hazard (thermal runaway in a Class C cargo compartment), an exposure (every flight the product takes), and a severity that, for any pack above the FAA's "small fire reasonably containable by halon" envelope, is Catastrophic. That hazard row should have been written before anyone shipped a unit. It wasn't.

3. A worked snippet

3.1 Missing operating-mode row (humanoid item definition, transport mode excerpt)

| ID | Operating mode | Environment | Energy state | Actuation state | Standards in scope | |----|----------------|-------------|--------------|-----------------|--------------------| | OM-08 | Cabin transport, stowed at passenger seat | Cabin altitude up to 8,000 ft, 15–30 °C, 30–60 %RH, vibration per RTCA DO-160G §8 Cat S | Battery installed, SoC bounded to 30–50 % per IATA DGR PI 967 Section II | High-voltage rail disabled; actuators de-energized; safe-state latched | 49 CFR 175.10(a)(18); IATA DGR 2026 PI 967 Sec II; ICAO TI Doc 9284; UN 38.3 (cell-level inherited) | | OM-09 | Cargo transport, Class C cargo compartment | Per RTCA DO-160G §4 Cat B2, §7 Cat B | Battery may be at shipping SoC; pack monitored | Powered off; cell-tap monitoring optional | 49 CFR 173.185; UN 3481 PI 967; IATA DGR 2026; Halon 1301 fire-suppression compatible packaging | | OM-10 | TSA checkpoint, X-ray and ETD screening | 15–30 °C, brief mechanical handling | Battery installed at carry SoC | Locked-out / tagged-out; tamper-evident chassis | TSA SD-1542-04-08H; 49 CFR 1544 |

The point of this excerpt is the boring one: every other operating mode in a humanoid item definition gets an environment, an energy state, an actuation state, and a standards-in-scope column. The transport modes don't, because the team building the robot has never been asked to write them. Until they are, no derived requirement can trace to them.

3.2 Hazard row — transport mode thermal runaway

| ID | Operating mode | Hazard | Cause | Effect | Severity | Mitigation | |----|----------------|--------|-------|--------|----------|------------| | HZ-TR-01 | OM-08 (cabin stowed) | Thermal runaway of installed Li-ion pack during cruise | Installed pack energy above 100 Wh threshold, no operator approval, no IATA DGR Class 9 declaration; pack lacks UN 38.3 re-qualification at assembly level | Cabin fire, smoke, possible diversion or loss of aircraft | Catastrophic | Constrain installed pack to under 100 Wh; provide removable spare in the 100–160 Wh band, two-per-traveler max, individually short-circuit protected per 49 CFR 175.10(a)(18); print UN number, Wh rating, and proper shipping name on chassis; require operator approval flow for any pack between 100 Wh and 160 Wh |

Note the rule for prose: the regulation says "under 100 Wh," not "< 100 Wh." If you write that hazard with the math symbol the MDX compiler will rightly assume you opened a JSX tag, and the build will die. Same engineering discipline as a header on a steel coupon: respect the format you are working in.

3.3 Compact FMEA — transport-mode battery subsystem

| Item / function | Failure mode | Cause | Local effect | System effect | S | O | D | AP | Action | |-----------------|--------------|-------|--------------|---------------|---|---|---|----|--------| | Battery pack (cabin transport) | Energy exceeds 100 Wh, no operator approval | Pack designed for endurance, no transport-mode SKU | Gate refusal | Flight delay, customer harm, regulatory exposure | 7 | 8 | 2 | H | Add 95-Wh "transport-mode" pack SKU; physical interlock between transport-mode and ops-mode packs | | Battery management system | Fails to assert "transport safe" before chassis is handed to a courier | No supervisor state for transport; SoC not bounded to 30–50 % | High SoC in shipping | Thermal runaway risk amplified | 9 | 3 | 4 | H | Add explicit transport-safe supervisor state with cell-balance, SoC bound, and high-voltage isolation; require state on chassis e-label before shipping label is printed | | Chassis labeling | Missing UN 3481 / UN 3480 marking, missing Wh rating | No labeling requirement in product spec | Hazmat handlers cannot classify | Refusal, fines, possible diversion | 6 | 6 | 3 | M | Add IATA-compliant battery label and Wh rating to chassis; require sign-off in Production Part Approval | | Power architecture | Actuators energized at security screening | Sleep mode optional, not enforced | Unexpected motion in checkpoint | Injury risk, screening refusal | 8 | 4 | 3 | M | Require hardware-latched safe-state in transport mode; no software-only de-energization |

S/O/D rated 1–10 per AIAG-VDA 2019; Action Priority H/M/L per the AIAG-VDA lookup table. Two High-AP rows in a four-row mini-FMEA is, in audit terms, a bad week.

4. Derived requirements (excerpt)

These are the kind of requirements I would expect to find in any humanoid-robot product file before the first unit ships. They trace cleanly to 49 CFR 175.10, IATA DGR PI 965–970, UN 38.3, and ISO 13482's scope clauses.

REQ-TR-001 — Installed pack energy. The installed primary battery pack, in any unit intended for cabin transport, shall not exceed a Watt-hour rating of 100 Wh, computed per 49 CFR 175.10(a)(18)(i) and IEC 62133-2 nameplate convention. (Traces to 49 CFR 175.10(a)(18)(i); IATA DGR 2026 PI 967 Sec II.)
REQ-TR-002 — Spare pack envelope. Where additional endurance is required beyond a 100-Wh installed pack, the product shall be designed to accept up to two removable spare batteries in the 100–160 Wh band, each individually short-circuit protected by hard-shell case, terminal cap, or original retail packaging. (Traces to 49 CFR 175.10(a)(18)(i)(B); IATA DGR 2026 PI 970 Sec II.)
REQ-TR-003 — Transport-safe supervisor state. The BMS shall expose a "transport-safe" state in which: (a) high-voltage rail is hardware-isolated; (b) actuator power is removed; (c) State-of-Charge is bounded to 30–50 %; (d) state entry is non-volatile and tamper-evident on the chassis e-label. Time to enter the state from any operating mode shall not exceed 10 seconds. (Traces to ISO 13482 §5.7, IATA DGR 2026 PI 967 Sec II.)
REQ-TR-004 — Chassis hazmat label. Each unit shall carry a permanent, machine-readable label declaring battery chemistry, Watt-hour rating, UN proper shipping name (UN 3481 if installed, UN 3480 if shipped separately), and a transport classification per 49 CFR 173.185. (Traces to 49 CFR 172.101; IATA DGR 2026 §7.)
REQ-TR-005 — Pack-level UN 38.3 evidence. A UN 38.3 (T1–T8) test summary shall be on file for each unique battery-pack assembly, not only for the underlying cells, and shall be incorporated into the product technical file per ISO 13482 §6.1 and ANSI/A3 R15.06-2025 risk-assessment evidence. (Traces to UN Manual of Tests and Criteria, 7th rev., Section 38.3; ISO 13482:2014 §6.1.)

None of these are exotic. They are the kind of rows that, in an automotive program, would have closed at the item-definition gate and never made it to a NHTSA recall page.

5. What the headline really tells us

The story being told publicly is "humanoid robot bought a plane ticket and Southwest freaked out." The story I read is duller and more useful.

Two product teams shipped a humanoid robot with no transport-mode operating state, no hazmat conformance flowdown, and no chassis label that a TSA officer or a gate agent could read. The fact that Stewie's owner had to swap to a smaller battery to make screening work, and the fact that Bebop's pack was removed at the gate and put in checked compartments under God knows what packaging, are both consequences of the same missing artifact: a transport-mode hazard analysis that should have been written when the product line was scoped.

This is solvable. It is not a robotics-research problem. It is a half-day workshop with a robotics safety engineer, a logistics specialist, and someone who has actually filed a UN 3481 Dangerous Goods Declaration. The output is one operating-mode table, one hazard row, one FMEA pass, and five derived requirements. The deliverable is roughly the depth of what I wrote in the section above. The cost of not writing it is, at minimum, an hour of delay in Oakland and, at maximum, a Class 9 incident in a Class C cargo compartment at 35,000 feet.

The headline is the symptom. The missing transport-mode artifact is the disease. The cure is unglamorous, well-understood, and already sitting in the standards on the shelf. We just have to write it down.

Sources

— Jherrod Thomas, The Lion of Functional Safety™