Learning in lane-level route planner

An apparatus for lane-level route planning, comprising:

• a processor configured to:
• obtain lane-level information of a road,
• wherein the road comprises a first lane and a second lane, and
• wherein the lane-level information comprises first lane information related to the first lane and second lane information related to the second lane;
• convert the lane-level information to a state transition function;
• obtain a policy as a solution to a model that uses the state transition function,
• wherein, given that a vehicle is on a first lane segment, the policy provides an action for controlling the vehicle with respect to moving to a second lane segment that neighbors the first lane segment,
• wherein the model incorporates, into a cost function, a user preference for objectives including at least a first objective relating to comfort and a second objective relating to urban lane segments, and a user input indicating a slack time for alternative routes to a destination relative to the user preference,
• wherein the cost function represents an expected immediate cost of performing the action in a state,
• wherein the first objective relating to comfort indicates a preference for lane segments where traffic moves at speeds that are within a threshold speed of speed limits on the lane segments,
• wherein the second objective relating to urban lane segments indicates whether to maximize selection of lane segments classified as urban, and
• wherein the slack time indicates an allowable deviation in travel time relative to the user preference;
• receive the destination over lane segments that are not classified as urban;
• and
• at time steps along a route to the destination, autonomously control an AV according to one or more respective actions obtained from the policy based on respective current lanes of the AV.

A non-transitory computer-readable storage medium, comprising executable instructions that, when executed by a processor, facilitate performance of operations comprising operations to:

• receive from vehicles traversing a road respective lane-level information;
• obtain a navigation map from lane-level information received from the vehicles, the navigation map comprising a state transition graph and a reward function, wherein the lane-level information comprises at least two of environment information, vehicle information, or human information;
• transmit the navigation map to a vehicle, wherein the vehicle uses the navigation map to obtain lane-level routes;
• obtain a policy as a solution to a model that uses the state transition function,
• wherein, given that a vehicle is on a first lane segment, the policy provides an action for controlling the vehicle with respect to moving to a second lane segment that neighbors the first lane segment,
• wherein the model incorporates, into a cost function, a user preference for objectives including a first objective relating to urban lane segments and a second objective relating to comfort, and a user input indicating a slack time for alternative routes to a destination relative to the user preference,
• wherein the cost function represents an expected immediate cost of performing the action in a state,
• wherein the first objective relating to urban lane segments indicates whether to maximize selection of lane segments classified as urban over lane segments that are not classified as urban,
• wherein the second objective relating to comfort indicates a preference for lane segments where traffic moves at speeds that are within a threshold speed of speed limits on the lane segments, and
• wherein the slack time indicates an allowable deviation in travel time relative to the user preference;
• receive the destination;
• and
• at time steps along the lane-level routes to the destination, autonomously control an AV according to one or more respective actions obtained from the policy based on respective current lanes of the AV.