We are focusing our efforts towards an on-demand highway autonomy solution. We believe it to be an instance of level 4 autonomy that is achievable in significantly less time, risk and resources than many of the other level 4 approaches, yet still enabling serious opportunities in both the trucking and personal transportation industries.

By on-demand highway autonomy we specifically mean two things:

• Our solution is limited to highways, and more precisely European highways which are extremely standardized and regulated.
• Our solution is only available on-demand, on pre-defined zones covered by the network, if general conditions allow it.

The two classes of products we can build on top of this technology are:

• An on-demand highway autonomy network that can be trigger by human drivers already driving on highway (consumer market mostly).
• Fully autonomous vehicles on fixed routes with remote control for entering and exiting the highway domain (trucking).

We target level 4 autonomy with no disengagement requirements within a proposed on-demand autonomy allocation. The vehicle shall safely stop if needed in case a human is not available at the end of the allocated autonomy period or progress is prevented within the autonomy zone due to unforeseen conditions.

European highways benefit from uninterrupted emergency stopping lanes and on-road safety parking zones every 2km enabling us to fully operate on the highway with no need to manage an exit out of the regulated highway space (not even a resting area that would require handling an exit ramp). If the driver (inside vehicle or remote) fails to re-engage at the end of an on-demand autonomy zone, we can safely stop and wait on a safety parking zone. Similarly, if an exceptional situation prevents forward progress within the autonomy zone we can secure the vehicle on available emergency stopping lanes.

Our approach is designed to limit the scope of our autonomous agent to the regulated highway space which triggers in turn a chain of simplifying assumptions that we expose below.

Symbolic Highway Representation

Because of their nature, highways can be represented symbolically in a much simpler manner than road networks.

European highways don't have intersections, only standardized merge-outs and standardized merge-ins. As such, vehicles in interaction in that environment are all going in the same direction at all time, on parallel lanes, along a curved trajectory; there is no valid motion on highways that makes you go backward or even orthogonally to the lanes direction. For that reason, the output of motion planning in that environment is very coarse as it solely consists in a window of future target lateral positions and target forward speeds. Additionally it exists a canonical bijection (given a prior map) from the real-world coordinates to a symbolic rectilinear space compatible with the nature of the motion planning output, also called lane-centric coordinates. We propose to execute motion planning within such a coordinate space (and propose a precise definition later on), mapping back to real world coordinates, once the planning is done, at time of vehicle control.