Current ventures creating self-driving cars and trucks have everything wrong. Every automaker and technology company is focusing solely on the individual car rather than the potential colony that they can make. Rather than focus on throwing radar, lidar, and lasers into supremely equipped vehicles, the market players should begin to think about the long-term systems that should be used and standardized, just as NIST establishes standards for Technology.
Vehicle Automation and the Wrong Approaches
The current market is made up on dozens, if not hundreds, of technology companies and automotive manufacturers trying to develop individual self-driving technologies that will be suited to go into each of their own cars, with their own patents, and their own ability to sue each other to gain market share. The primary problem is the fact that these companies are all getting away from their core business offerings, and investing time, money, and nonrenewable resources into variations of the same failing concepts.
The Technology companies, such as Google and Waymo have taken vehicles that are currently in production and have outfitted them with sensors (see radar, lidar, laser, and cameras) and software where the individual car and travel on a pre-determined waypoint system. Each vehicle is heavily and painstakingly modified and tested to ensure NHTSA safety standards. Here’s the problem with the likes of Google and Waymo: They have truckloads of cash and technology at their disposal, but they are not automotive experts and their business model is all wrong.Google owns entire server farms and commands a 35% share of the globe’s advertising revenue. Google is, by and large, a technology company with a lot of software and servers (much of which I personally like and use), but they have zero experience in developing software for mass-production vehicles. Although, it would be nice for an Android-based platform to be developed for certain legacy automobiles.
Ford and Volkswagen have entered into an alliance to develop autonomous vehicles with a $4 billion investment stake by VW into Ford’s startup Argo AI. The problem with automakers delving into billion-dollar equity investments in each other’s autonomous vehicle divisions is automotive companies, like the aforementioned benchmarks, are notorious for developing technology and guarding their patented technologies with a vengeance. See Robert Kearns. The problem with these aggressive corporate tactics is although it may ensure the dominance of market share for the manufacturer who owns the technology, the price-point for the mass market will always be significantly higher if the technology cannot be scaled across the entire industry.
Enter Tesla, Inc. a hybrid between automotive manufacturing and technology companies. Their mission is to produce mass-produce emissions-free Electric Vehicles (e.g. Tesla Models S/3/X/Y) that are marketed with the Tesla Autopilot system. Tesla’s “Autopilot advanced safety and convenience features are designed to assist drivers with the most burdensome parts of driving. All new Tesla cars come standard with driver assistance features such as emergency braking, collision warning and blind-spot monitoring.” Teslas are also available with Full Self-Driving Capability with limited functionality until software updates can be fully proven and rolled out with Over-the-Air (OTA) software updates.
The problem with Tesla is even the base version Model 3, without Autopilot or Full Self-Driving Capability is still too expensive for the average consumer. For the purposes of this discussion, a family of five with a gross annual income of $130,000 is the average consumer. The Autopilot system is an extra $3,000 and the Full Self-Driving Capability (an add-on after Autopilot) is another $5,000. It would take nearly 13 years to reach a break-even-point between the Model 3 and a 2014 Ford Explorer Limited. Tesla’s vehicles are extremely clean and simple machines. While the technology underneath the chassis is not anywhere close to simple, Tesla’s powertrains have significantly fewer moving parts, and they are heavily technology-laden. This can be a great feature for the future of Tesla vehicles, the global automotive market, and the mass-adoption of self-driving vehicles in new and used models alike.
What all these companies have in common is their drive to pack as much hardware and software into their own cars, using their own standards, their own technologies, and their own systems that incur a cost that the average consumer does not want to bear. Automotive prices are already increasing without the inclusion of technology and the US national automotive debt hit $1.274 Trillion in the 4th Quarter of 2018. The increase in debt, and the 6.7 year life-cycle of the average ownership experience increases the risk of a significant correction in the global automotive market. China is currently experiencing a significant market correction with a 15% Year-on-Year drop in sales. So the prospect of consumers paying more for higher-priced vehicles is reaching a break-even point. The average family needs to get from work to home during a round-trip commute. What is it worth to the average family in paying more for the same?
What is the price of lacking standards? The automotive industry has selected the National Highway Transportation Safety Administration (NHTSA) to serve as the organization responsible for setting the standards for self-driving vehicles. The problem is the NHTSA is great at setting automotive standards, but information technology is not their industry. Information Technology is in the backyard of the National Institute of Standards and Technology (NIST). Automotive competitors and Information Technology companies need to reconcile their technology with one another and set standards that conform to the hybridization of automobiles and information technology.
Availability vs. Possibility
At the rate current developments are going, there will be a multitude of vehicle platforms that are delivered with vehicle automation as a standard feature. The price, of course, will be borne by the consumer. However, one can posit that the dozens to hundreds of different variations in hardware and software configurations delivered with these vehicles will eventually cause interference with one another and do little to achieve their alleged goals of reducing driver fatalities. The development time and investment required means a loss of profits to shareholders, and an extended timeline within which a fully-autonomous vehicle is truly achieved.
SAE International established 6 levels of Driving Automation:
- Level 0: Current Market – System issues warnings, but has no control of vehicle
- Level 1: Hands-on – The driver and the system share control of the vehicle. (i.e. Adaptive Cruise Control and Parking Assist)
- Level 2: Hands Off – The automated system takes full control the vehicle in terms of acceleration, steering, and braking, but the driver must still monitor the road.
- Level 3: Eyes Off – The driver can safely turn their attention away from driving tasks, but must be prepared to intervene when necessary.
- Level 4: Mind Off – The driver can safely avert all attention from the task of driving as the system is in full control and is capable of navigating all obstacles and pathways
- Level 5: Steering Wheel Optional – No human intervention is required at all.
In July 2015, Andy Greenberg wrote an article for Wired titled, “Hackers Remotely Kill a Jeep on the Highway – With me in it.” Andy’s friends, Charlie Miller and Chris Valasek had hacked Andy’s Jeep and were able to fully operate every system in the Jeep, going so far as to bring him to a full stop from 70 mph and adjust the air vents on the interstate. The achieved this by hacking the vehicle’s communication system using a laptop while in their apartment. This prompted Fiat Chrysler (FCA) to issue as security patch to the Zero Day hack on their vehicles.
But while FCA saw the Zero Day exploit as a hack that could damage their reputation and stock price, they missed an opportunity to capitalize on the technology they already have in place and becoming the driving force pushing the limits of Level 5 autonomy. There are still manufacturers who do not have the scope of technology in their vehicles that FCA did in the Jeep back in 2015, but there is an entire industry of aftermarket devices that can enable a computer to remotely control a vehicle.
Taking the idea of the 2015 Jeep Zero Day Hack, and applying the principles of drone automation, and data center artificial intelligence architectures, M2 Dynamics is developing a network architecture that will hack the hive mentality of a colony of bees to achieve Level 5 autonomy. The methodology used is to reduce the quantity of hardware and software currently being tested in autonomous vehicles and utilizing a central data center to communicate with each car as a collective. Think SkyNet without a rampaging self-aware AI program.
The central idea is to connect all vehicles that have the current capability to be controlled by the vehicle’s computer. The vehicle’s computer must be able to control braking, steering, acceleration, as well as transmission functions. The central system, called HiveNet, will be connected to each vehicle in its network through a dongle attached to the vehicle’s OBD II system. The vehicle’s OBD II dongle and HiveNet will actively communicate with each other and help the community of vehicles navigate through a concept called Spatial Memory, much like how Honey Bees navigate, as studied by Randolf Menzel, et. al.
Spatial Memory has multiple representations that help bees and humans alike develop cognitive maps to help them navigate the world around them.
Short-Term Spatial Memory (STM) is the temporary storage and management of information that is necessary to complete complex cognitive tasks such as learning, reasoning, and comprehension. As it relates to HiveNet and the connected vehicle, STM would be the HiveNet talking to the car and the other cars in the vicinity, providing proximity information, vehicle speed, and even intended route. STM also addresses when a road is under construction for a period of time (i.e. a month or more).
Spatial Working Memory (WM) is much like a computer’s RAM in which it only temporarily stores information to navigate a situation. For instance, the your car will encounter other cars in the immediate area, and your car will use its WM to maintain speed and distance from other vehicles. Once the other vehicles leave your car’s proximity, the WM is dumped and the process restarts. This representation is already in use by Google Maps in the case when a user is navigating from home to work and Google Maps provides live detail of the traffic conditions (i.e. Green, Yellow, Orange, and Red Zones). WM addresses whether there is an accident ahead and if a police officer or tow-truck driver are diverting traffic from two lanes to a single lane.
Long-Term Spatial Memory (LTM) is such a representation like a map or 3D model of the vehicle’s surroundings and intended route that may be updated once every month or so, depending on if there is any construction. LTM will include road maps, scans of existing buildings and bridges, and traffic signals. Civil Maps has introduced edge-based HD mapping and localization to enable fully autonomy anywhere. Its Edge Mapping ™ technology enables self-driving cars to “build, share, and update a mental model of the environment while continuously determining their location.” This solution also ties in very well with STM and WM.
Figure 1 Illustrates the relationship between HiveNet and Vehicles on the road. HiveNet is an application powered by Artificial Intelligence and serves as the central node of information. It knows where each vehicle is, where they’re going, and what route they’re taking to get there. More importantly, it communicates to the car directly and controls their path to destination. HiveNet and the vehicle will communicate through a mixture of network pathways, including fiber and cell service. The vehicle’s OBD II system, derived from the OpenXC™ Platform, developed by Ford Motor Company researchers.
While the system proposed by M2 Dynamics will not utilize the exact dongle as shown in Figure 2, the general idea behind the use of CAN Bus communications on OBD pins is the same. The CAN Bus dongle will include an LTE module with future devices utilizing available 5G networks. In using the various memory representations, we can limit the amount of data that is required to be transmitted between the end-user vehicle and HiveNet, eliminating the need to transmit and receive Gigabytes of data. The intent is to ensure a safely-controlled autonomous vehicle is able to get from origin to destination with as little data as possible.
The development of the system across multiple vehicle platforms will require a community of developers, much like XDA Developers and their support provided to the development of Android-based operating systems such as LineageOS across a range of phone manufacturers. The primary framework will be locked from manipulation, but developers will be able to submit updates through a review process to ensure the integrity and security of the software.
The utilization of code review for the vehicle-side software and the implementation of the representations of spatial memory will ensure that while the systems and nodes will interact, they will not be able to be manipulated by hostile actors. Control of the solution would have to be maintained by an organization that is neither government nor private, incapable of being beholden to markets and investors. The corporate framework would necessarily be that of a public-private partnership where the system is developed, controlled and maintained by a non-governmental organization, but potentially supported by funds from the respective local and state governments.
Investors would not likely benefit from a public offering, but could achieve passive return on investment. For instance, a “good neighbor” insurance company could decide to make a grant to the entity and the return would be fewer claims. A telecommunications giant would be able to provide a grant, but would only receive a return on the SIM cards activated on its networks through the OBD II CAN Bus dongle. As personal data security would be paramount, all information passing between the vehicle and HiveNet would be secured utilizing an encryption method that can only be decrypted by HiveNet and the dongle. No raw, data will be made available to any entity outside of the HiveNet system and the end-user.
The current iterations of OpenXC ™ hardware range from $130 to $425, depending on features and functionality. The investment between Ford and Volkswagen of $4 billion is equivalent to purchasing and deploying over 30.7 million systems to vehicles. In 2018, Ford sold just under 2.7 million vehicles in the US, the equivalent of spending $1,481 per vehicle on equipment to get a vehicle ready for automation.
M2 Dynamics, LLC is working on developing a solution called HiveNet that will take the approach of utilizing extra-vehicular networks and memory representations to control a vehicle and the vehicles around it rather than loading up a standard vehicle with thousands of dollars worth of equipment, pushing even a base-price vehicle into luxury territory. In securely networking vehicles, HiveNet will be able to accelerate the deployment of a fully-automated experience with limited investment and significantly less development time.