At some point, every self-driving car needs the skills of a F1 driver
Canopy Co-Pilot : the missing piece of self-driving technology the world has been waiting for
In our recent white paper (available on request) we discussed a new project for Canopy Simulations, that of turning the engine behind Dynamic Lap into a controller for a self-driving car. If you’re interested in this idea then please do get in touch and we’ll send you a copy of the white paper, the outline is as follows:
Motorsport simulations do more than just allow the car to be driven in a virtual environment, they actually drive the car. Not only do they drive the car, but they drive it like a racing driver would, in other words they drive it at the limit of grip, as fast as possible. Most simulations do this in a very loose sense, using several unrealistic assumptions to simplify the problem. Canopy’s simulations use a technique called ‘collocation optimal control’, or simply ‘collocation’ to drive the car without making these assumptions. The result is that the car is driven to produce the minimum lap-time whilst following a trajectory which could actually be driven in real life. Not only that but collocation simulations can do this whilst optimizing for other things too, like optimal energy usage, or optimal tyre conservation (see white paper for full details). Motorsport simulation, specifically Canopy’s collocation-based simulations, are the only place where a fully dynamic model of the car is optimally controlled at the limit of grip. When I say the ‘limit of grip’ I mean in situations where the car is one wrong move away from sliding out of control, or spinning.
Cars don’t usually operate at the limit of grip, in fact safe drivers on the road will take steps to make sure that they always have some grip to spare so that they stay away from the limit (e.g., slowing down and leaving more space when it’s raining). However, when something unexpected happens and evasive action is required, there is often no choice but to operate at the limit. Emergency braking is a simple example, if a child steps out in front of you, you hit the brakes and try to stop as soon as possible. If you hit the child without reaching the limit of grip first, then you could have tried harder! This is the case with all emergency manoeuvres, if you can’t control the car at the limit of grip then you will have some collisions that could have been avoided. Up to this point autonomous vehicles have been operating well within the limit of grip; on dry roads, at low speeds such as in a city or suburban environment. Where speeds are higher the environment has been very benign e.g., highways/motorways. However, most fatalities occur in rural areas, where lanes are uneven, surfaces are poor (or covered in manure) and surprise obstructions are common (trees, oncoming cars, animals). Level 5 autonomous cars will have to deal with all of these, and as Level 4 cars expand their envelope, they will encounter more and more situations where an emergency manoeuvre is required and where that manoeuvre involves the car sliding in some capacity over the road. These situations cannot be handled by simple controllers that are designed for low speed, high grip situations.
Whilst the problems of localization, perception, object identification etc. have received a lot of development effort in the world of self-driving cars, this problem of control in unusual situations has received very little attention. Canopy is now addressing that deficit by applying our expertise in collocation optimal control to work in real time on a car.
Co-Pilot will repeatedly predict and plan the path and controls for the car up to the horizon
The resulting collocation pilot, or Co-Pilot, will repeatedly predict and plan the path and controls for the car up to the horizon (as far as the car can ‘see’). After executing the first step of the planned path, Co-Pilot will compare what actually happened to the plan and adjust the parameters of the model to best account for the difference (maybe a change in grip, or a revised car mass etc.). This cycle of learning and re-planning with new information will happen many times every second that the car is driving along.
Co-Pilot will enable autonomous vehicles to handle themselves expertly even in the most extreme situations of low grip and high-speed evasive manoeuvres. Not only that, but using Co-Pilot to plan for multiple what-if scenarios will enable action to be taken earlier to avoid getting into an emergency manoeuvre in the first place. Finally, the presence of Canopy’s collocation solver on board the car will bring with it all of the benefits that our motorsport customers currently use it for. These benefits include optimal use of a hybrid powertrain to use minimum energy for a journey, management of battery usage and temperature to prolong life, optimal control of multiple motors for maximum efficiency, etc. etc. In addition to these transferable benefits, road vehicles will also benefit from control optimization for passenger comfort. This will make the autonomous vehicle feel safer as well as actually being safer – this could be a major differentiator in the uptake of self-driving cars.
Canopy’s world-leading expertise in using collocation to control cars at the limit of grip is being applied to self-driving cars to fill a gap in present autonomous technology and enable the expansion of the Level 4 envelope and the roll out of true Level 5 driving.
If you’d like more details then we will be more than happy to provide a copy of our white paper, explaining the project in much more detail. We are very much looking forward to showing our technology to potential partners and discussing integrations and collaborations.