Deadlock prediction for traffic planning

1 . A computer-implemented traffic planner for planning a plurality of guided vehicle trajectories for a plurality of heavy-duty vehicles to follow within a defined region, wherein the vehicle trajectories pass through an area where vehicle occupancy is constrained more than it is elsewhere along the vehicle trajectories in the defined region, each vehicle trajectory comprising a vehicle path and a speed and direction of travel for a heavy-duty vehicle to travel along that vehicle path, the traffic planner comprising:

a deadlock classifier model trained to predict whether or not an initial non-deadlock traffic situation, x comprising the plurality of heavy-duty vehicles having initial locations in the defined region at an initial planning time to at the start of a first planning time period will lead, in a later planning time period t within a planning horizon, to a subsequent traffic situation comprising a deadlock traffic situation, wherein each input traffic situation, x , represents trajectory segment vehicle occupancies at the initial time to for vehicles following the plurality of guided vehicle trajectories within the defined region, and wherein each subsequent traffic situation represents different trajectory segment vehicle occupancies at subsequent planning time periods t by the plurality of heavy-duty vehicles following the plurality of guided vehicle trajectories;

wherein the deadlock classifier model is configured to output a prediction of the subsequent traffic situation deadlock classification up to the planning horizon based on the input traffic situation representation, x , at the start of the planning horizon and

wherein the traffic planner is configured, based on a received traffic situation x for a plurality of heavy-duty vehicles following guided vehicle trajectories at a time to being input to the deadlock classifier resulting in a prediction of a deadlock traffic situation at one or more subsequent times within the planning horizon, to update the subsequent vehicle actions for at least one of the plurality of heavy-duty vehicles to perform along the path of its guided vehicle trajectory.

2 . The traffic planner of any claim 1 , wherein the deadlock classifier model was trained using a training data set generated by a traffic situation representation generator for the predetermined region configured to generate, for a plurality of vehicles following vehicle trajectories within the predetermined region, a binary traffic situation representation of the predetermined area comprising, for each of the plurality of vehicles, an initial vehicle trajectory segment location in the predetermined area and a subsequent deadlock status.

3 . The traffic planner according to claim 1 , wherein a deadlock classification represents two or more vehicles being unable to advance to their next segment locations along their trajectories in the one or more subsequent time periods.

4 . The traffic planner according to claim 3 , wherein the traffic planner plans, at each of the one or more subsequent time periods, for each of the plurality of heavy-duty vehicles in the defined region, a vehicle trajectory along which that vehicle either advances to a next segment location along its trajectory path or waits at its current segment location on its trajectory path.

5 . The traffic planner according to claim 4 , wherein each of the plurality of heavy-duty vehicles follow the same path along their guided vehicle trajectories and wherein each of the plurality of heavy-duty vehicles occupies a different segment along the path at any given point in time.

6 . The traffic planner of according to claim 1 , wherein the number of planning time periods from the start of the initial planning time period to the planning horizon is more than one and the vehicle trajectories are adjusted dynamically in real-time.

7 . The traffic planner according to claim 1 , wherein the heavy-duty vehicles are autonomous heavy-duty vehicles following predetermined guided vehicle trajectories in the predetermined region.

8 . The traffic planner according to claim 1 , wherein the vehicle occupancy constrained area within the defined region through which the plurality of vehicle trajectories pass comprises a one-way section of a route along which at least two vehicle trajectories pass in different directions along their trajectory paths.

9 . A method of determining if a vehicle traffic situation comprising a plurality of heavy-duty vehicles following a plurality of guided vehicle trajectories in a defined area will lead to a vehicle deadlock, the method comprising:

generating a traffic situation representation of the defined area comprising, for each of the plurality of vehicles, an initial vehicle trajectory segment location and a subsequent deadlock status;

inputting the traffic situation representation to a traffic deadline classifier trained to output a traffic deadlock classification based on an input traffic situation representation of traffic in the area; and

based on the input traffic situation representation, generating a predicted deadlock classification of a subsequent traffic situation in the defined area for that input traffic situation representation.

10 . The method of claim 9 , wherein the method is implemented using a neural network configured to receive as an input vector a binary string representation of a traffic situation, wherein binary string comprises a plurality of elements where a vehicle grid-lock may occur, each element representing an occupancy state of a traffic segment along one or more vehicle trajectories in the defined area.

11 . The method of claim 9 , wherein the traffic deadlock classifier is modelled using one of:

a feed forward neural network;

a look-up table;

a support vector machine;

a radial basis model.