Scaling up the Robotaxi with reliable rides

The winner is...

And so... after much fanfare, the winner of "the first self-driving car in commercial use" will go to the aptly named robo-taxi. A shuttle bus or a ride-hail car without the driver.

Alas, self-driving for every one of us in our own cars will have to wait a bit longer. For example, Ford and VW have shut down their self-driving initiative for passenger cars. VW Group put a halt to its Audi brands autonomous driving program to focus on its software platform roadmap. Tesla's "Full Self Driving" (FSD) beta is under scrutiny, well because the term "self-driving" has become vernacular and is used inconsistently. It has yet to be agreed whether Tesla FSD is partially, conditionally or highly automated driving system. Go figure.

Anyway, without going into the rise of and raison d'être of robo-taxis, there are about 38?US states that have enacted legislation or executive orders for Autonomous Vehicles (AVs). California and Nevada have legislation that authorizes the full deployment of AVs without a human operator. A so called “Level 4” highly automatic driving system is so sophisticated that it can reason about the road condition to adjust handling, evaluate the likelihood of a pedestrian jay walking or the probability of a forward collision unfolding milliseconds into the future. It has the memory to plan for short-term and long-term actions that adhere to traffic laws. That is impressive.

Robo-taxi brand wars

You would not know it, but the market has started to get crowded with real-world Level 4 robo-taxi companies that are operating in prescribed zones with necessary city permits.

Each company is charging forward to add cities as a signal they are successfully starting to expand markets (and soon after profitable growth):

1. Baidu's robo-taxi service, under its Apollo program has completed 1.4 million trips across 10 cities in China. Apollo Go has over 40 million kilometers of Level 4 autonomous driving test mileage. For the year 2023, Baidu plans to have 200 more fully driverless autonomous vehicles after receiving approval to operate without without safety operators as a backup.

1. Pony.ai, backed by Toyota Motor Corp has been issued a fully driverless autonomous vehicle road test permit in Beijing with plans to deploy ten driverless robo-taxis for testing in ‘challenging urban traffic scenarios’ and a safety officer will monitor the test vehicles remotely.

3. GM's Cruise has rob-taxi service in Austin, Phoenix and San Francisco relying on Level 4 equipped Chevrolet Bolt vehicles. The COO Cruise says the business has plans to enter 'a large number of markets' in 2023 and scale operations up to “thousands of vehicles” using a “repeatable playbook” developed in San Francisco, Austin and Phoenix.

4. Waymo started offering it's "rider only" service in August of 2022 in neighborhoods of San Francisco. A similar service already existed in Pheonix, Arizona since since May 2022 . In November 2022, Waymo?received a permit ?from the California Public Utilities Commission to offer fully autonomous vehicle rides to the public. It has been working with modified Jaguar I-Pace SUV’s and Chryster Pacifica Hybrid vans.

5. Zoox. A start-up that is now part of Amazon is on its journey to launch its unique brand of robo-taxi experience with the CEO saying that the focus is being on?public roads starting at a “small-to-moderate scale but putting in place an ability to add cities on a continual basis that's good for customers."

6. Motional is a joint venture between Hyundai and Aptiv for robot taxi services. The vehicle platform is Hyundai's Ioniq 5 uses which relies on more than 30 advanced sensors and onboard computing system to find its way in city streets. ?In November of 2022, Lyft and Motional launched a driverless ride-hail service in Las Vegas via the Lyft App. Uber is following suit with a 10-year deal with Motional with human safety operators in the initial stages in Las Vegas and LA but moving to fully driverless rides sometime in 2023.

7. May Mobility is a 5-year-old autonomous vehicle startup that has been testing its vehicles in Michigan, Minnesota and Texas---but all with human safety drivers. Originally designed as a fixed route shuttle, May Mobility is now positioning itself more about operating in "geo-designed areas " or zones that have been mapped extremely well.

It’s state-by-state regulations to operate

In the United States, every robo-taxi company has to abide by laws enacted in each state as well as take-up recommendations by organizations such as SAE (Society of Automotive Engineers) and the International Organization for Standardization (ISO). Specifically, the SAE and ISO put together a taxonomy (5 Levels of Driving Automation) that lays out what is expected as distinctly as possible for each level of driving automation. Each manufacturer then self-certifies their vehicle model's automatic driving feature within in a specific ODD (operational design domain).

For example :

In broken lane marking, where lane widths are >= 12 feet, in light rain, dark conditions, where the road service is dry and in rush hour traffic, where the operating speed is <35 mph, the vehicle CAN automatically "pass a double-parked vehicle that is blocking it's lane, where passing it requires crossing a double line lane to enter a lane of opposing traffic"

Here are some requirements extracted from CA regulations:

• There is a communication link between the vehicle and the remote operator to provide information on the vehicle's location and status and allow two-way communication
• That the manufacturer will continuously monitor the status of the vehicle and the two-way communication link
• How to communicate with a remote operator of the vehicle who is available at all times that the vehicle is in operation, including providing a contact telephone number for the manufacturer.

Here are some expectations of "Level 4" High Driving Automation features:

• A Level 4 ADS (Automatic Driving System) feature capable of performing the entire DDT (Dynamic Driving Task)!during valet parking (i.e., curb-to-door or vice versa) without any driver supervision.
• Level 4 ADS, however, will automatically perform the DDT fallback and achieve a minimal risk condition if the user fails to take over when the freeway ends (e.g., because s/he is sleeping).

Next up: Adequate supervision, control and orchestration of tens of thousands of robots

"...imagine it's a rainy day in San Francisco at 3:00 pm on a Tuesday in February. Most K-12 students are finishing classes for the day. The rain has increased the chance of an accident on a few steep roads that collect rainwater run-off. Traffic is starting to increase and will continue to until the rush hour peak each weekday at 5:50 pm." - Source: LinkedIn

How does the robo-taxi fleet respond to the changing conditions? Will there be a coordinated response to pre-position vehicles? What are the procedures and protocols for that day's unique journey? What happens in the case of an accident...will the vehicle be remotely parked by a tele-operator? What happens if the vehicle's diagnostics indicate the tire pressure is becoming a problem?

These questions are exacerbated when you have thousands of self-driving vehicles on the road—-owned by different operators.

Typically referred to as fleet management, the supervision, monitoring, control and diagnostics will be crucial to successfully scaling up operations. For example, Amazon Web Services' Fleetwise allows automakers, fleet operators and automotive suppliers to collect massive amounts of data including vehicle real-time diagnostics, GPS location and signals about the EV battery health in order to improve fleet operations.

Scaling-up of course goes well beyond fleet management. It is a business plan that starts with customer acquisition and effective financial and supply chain management. Scaling is what makes or breaks most businesses.

However, scaling up a robo-taxi enterprise will also mean being very thoughtful about the supervisory, control and orchestration capabilities of a fleet of robots. Robots that were purposely designed to make independent decisions. Lots of unintended consequences need to be well understood.

In turn the scope of traditional fleet management is already evolving to encompass capabilities from other industries that have been very successful themselves at scaling the operations of drones and robots:

Data Lakes & Digital Twins

In the industrial sectors there is proficiency in managing connected equipment at scale with real-time monitoring and control systems. For example, smart factories today have machinery sensors to predict maintenance, assembly line faults and other signals to ensure through-put is maintained. Every industrial company now has already taken a step further with the concepts of digital twins of the factory, it’s processes and equipment—while not applied everywhere it’s standard now to have some virtual copy of the telemetry that is sensed from a remote system and rendered either in real-time or via offline playback. The autonomous driving industry has adopted digital twins to model and control it’s “robotic systems” during the various stages of design and verification. Some are also automatically updating details of the virtual model based on real-world performance. More forward looking concepts call for higher fidelity "in-service" digital twins that are seamlessly integrated with the ADS —harvesting (and learning from) the collective memory of the fleet. For example, Tesla explained during its latest AI Day that it is mining its fleet for incorrect inferences to correct the labels and retrain the decision models ---all in a closed loop fashion. Similarly, Porsche has described a “big data loop” architecture for the life-long learning of its algorithms

Mission Control Systems

In the satellite and rocket launch arena a mission control system (MCS) is designed to receive, process and display data (telemetry), but more specifically send control commands to satellite during the LEOP (launch and early orbit) and routine phases. For example Spacelink has over 1700 satellites with the intention of deploying nearly 30k satellites. How do they track and operate such a network? What is the processes for in-orbit servicing? The automotive industry does have similar needs in mission-criticality. For example teleoperations and reliable connectivity and remote control capabilities are going to be necessary to assist ADS in the case of failure or uncertainty.

Feature Management for Software

Robo-Taxis have lots of software inside them which must be updated and configured on a regular basis. Now the domain of software configuration management, continuous integration and deployment (CI/CD) of code was practically invented by Silicon Valley style software companies. However for automakers it’s not that straightforward to just click “update” and have software in the vehicle be refreshed. There are various reasons for the difficulty: intermittent wireless connectivity, need to validate updates for safety-critical functions and discrepancies of hardware generations and software releases. One way around this has been to pre-install all “future” software on the vehicle. In that case the automotive industry can benefit by adopt “feature management” capabilities that allow them to toggle / turn on or off existing code after it has been deployed. The outcomes are better quality control and the ability to turn off features if there issues in the field. There would be no need to conduct roll-backs of software across a fleet of vehicles.

Cruise Robotaxis Roll Out in Tesla’s Home Town

What is Level 4 Autonomous Driving?

What is a SCADA System and How Does It Work?

What is a Computerized Maintenance Management System?