Might a failure of one infrastructure domain bring down an entire self-driving car fleet? A blackout caused by a substation fire, and a ripple effect that left more than 130,000 households without power, became a dramatic teaching opportunity over the weekend in San Francisco. While a blackout caused by a substation fire triggered by shots fired at 1:09 p.m., resulting in flickering lights and power outages affecting 21,000 households in an area bordering the Presidio to downtown San Francisco, Waymo’s fully driverless robot taxis, offering a glimpse of the future of transportation, had to stop operating – despite an unaffected Tesla-supervised ride service.
Waymo’s Jaguars were designed to have lidars, radar, cameras, and HD maps that were meant to mimic four-way stops at dysfunctional traffic signals. However, the magnitude of the blackout highlighted the underlying vulnerability in the system. Waymos had broken down at intersections in dozens of situations, hazard lights flashing as they obstructed bus traffic, prompting human drivers to find ways of circumventing the stranded vehicles. Videos emerged on social media showing five Waymos stalled at the same intersection, and a Waymo stopping in front of the city bus. According to Waymo, this connectivity issue may have had the added benefit of allowing the system to function without human administrators, reducing potential ambiguity in the system. “While the failure of the utility infrastructure was significant, we are committed to ensuring our technology adjusts to traffic flow during such events,” argued Waymo’s spokesperson.
However, this situation also brought out the complexity of highly autonomous fleets. The self-driving cars developed by Waymo do not operate as standalone cars but rather cloud-connected vehicles. These require traffic information through which they can function. When the connectivity is not available, “a safe mode is engaged and causes the robotaxis to stop.” Though this is perfectly fine for self-driving cars, but for “hundreds of robots” like those used by Waymo, this can be nothing short of disastrous. It’s just an “operational management failure” and warned, “If you have thousands of robotaxis that stop, you have a problem. What if this had been an earthquake?”
On the other hand, the support provided by Tesla is in the form of human drivers as well as Tesla’s “FSD (Supervised)” technology, which is based on camera vision and relies on human observation. Tesla’s California business was not affected by the connectivity aspect of the outage since they did not have the permit necessary to operate the service without human drivers. Tesla’s Elon Musk took advantage of the situation offered by the outage and tweeted that “Tesla Robotaxis were unaffected by the SF power outage.” However, according to Tesla’s own filings, they actually sought only the CPUC ride service permit to operate the human-driven service and not the six required to operate the autonomous service. As a technical issue, the blackout problem shows that there is a need for more robustness in self-driving cars.
The edge case problem, involving the condition of there being no traffic lights and performance under the conditions of degraded connectivity, is still prevalent. Those who participate in the MIT Mobility Forum, such as Bryan Reimer, propose a more holistic combination of human and technical intelligence, as against “an all-or-nothing world.” The example from the Toyota Research Institute’s Guardian system, which assists the driver in taking over the car in a case involving a lack of driver attention, retaining control as conditions permit, makes the point. It is called human-centric intelligent driving (HCID) and would prevent a “fleet paralysis” condition in the case of a failure in the infrastructure.
Infrastructure robustness is another region which is brought into play in this context. The more modern technology for lithium batteries could provide a definite safeguard against localized power failure. The use, as hinted in technical discussions associated withBattery technology, EV battery conversions already in use for solar and wind-based projects for power delivery on a local scale, already point towards this provision as a means for keeping both traffic management infrastructure and communication infrastructure operational under a power outage condition, resulting in interruption-free connection operations for a launched as well as a back-and-forth traversed connected self-driving cars and self-driving trucks via a web connection.
While self-driving cars and self-driving trucks are susceptible not only to accidental failure, malicious actions are also a danger for their favor. The safety record, as indicated by the strength denoted by the inputs into their system, which stand low 125 million miles as compared to human inputs is still an area it excels in, and it’s a proof indicator in itself that it has entered the urban setup, from the outskirts of London, London, and finally Phoenix, as it is satisfied with their system. But, as seen in the case at Waymo’s SF headquarters, a blackout problem serves as a warning that system robustness can have more layers than how reliable it performs while other conditions are normal as well. Waymo needs this realization, otherwise, it causes something more—immobility in innovation.
