Autonomous driving has been having its moment in the hype cycle for what feels like a while now. However, recent accidents and failed corporate promises have bought forward the question: How do self-driving cars really graduate from the research lab to the real world?
Mark Crawford, chief engineer of autonomous driving for Great Wall, shared his view of the current state of play, as well as the less talked about or new industries that will enable progress.
The majority of research is now concentrated on solving Level 4 autonomy. This is defined as a system robust enough to have both latitudinal and longitudinal control, plus be able to handle a single point of failure and get a vehicle into a minimal risk position without human intervention. Level 4 is operation within pre-determined design constraints such as geo-fencing, or day-time only. Level 5, complete automation – removing any constraints to operation – is a whole order of magnitude harder than this.
There are four major parts of the autonomous vehicle (AV) system (cue the acronyms):
First, the autonomous driving system (ADS). This function performs the sensing, perception and decision-making of the dynamic driving tasks to generate the motion control commands. AI plays a significant role here, including decision-making and mapping, with new applications being discovered all the time. This is the function, of course, that receives the majority of the press. ADS mainly comprises the sensors (of which Crawford says 360 LIDAR is closest to production level) and a combination of robotics and AI is required to reach Level 4/5 decision-making capability. The ADS interfaces with the vehicle actuation platform (VAP), which performs the actuation.
VAP is often overlooked. In fact, it is the foundation of AVs as it accepts the motion control commands to perform the vehicle actuation. It is by-wire control that evaluates the electric power steering, braking and power train systems. This still requires significant new research and engineering to achieve Level 4/5. Big challenges persist – achieving fail functional behaviour to address reasonable risks; incorporating manual control for operations; and addressing motion control for maintenance and service.
The ADS and VAP architecture link with the AV support system, comprising vehicle IT services (VIS) and vehicle management services (VMS).
These latter two are new industries and despite getting far less discussion, offer significant business opportunities. VIS manages the off-vehicle data and communications. Every vehicle will need to be connected in this way – this is the computers, software and wireless communication required to relay and retrieve information to and from AVs. It essentially helps the vehicle make a decision about its state to perform – for example, optimising a ride-pooling route or deciding when to return to the depot for refuelling.
The VMS keeps the vehicle operational and this sub-system is particularly in fleet-level systems. It aims to solve the questions of how to manage unmanned vehicles, how to optimise routine maintenance and when to conduct repairs.
The presentation was light on direct investment conclusions, but one point Crawford stressed is that collaboration will be key. Major parts will consistently be provided by multiple companies, but doesn’t restrict each supplier to niche components. For example, there seems to be a trend of one company developing the ADS and plugging into another’s VAP. In China for instance, he views Baidu’s open-sourcing of the Apollo architecture as a bold and game-changing move for the local market.
In terms of applications, he expects that the first genuine implementations will be in a captured fleet system focused on services such as ride-hailing or package delivery due to technology restraints and demographic demands.
This conclusion is also true in China, where adoption is expected to be advanced. Mobility as a service here is projected to be a US$2.5trn industry in China by 2030.