Multi-policy lane change assistance for vehicle
An advanced driver-assistance system (ADAS) comprises: a sensor; a behavior planner that performs multi-policy lane change assistance for a vehicle by evaluating multiple scenarios based on an output of the sensor using a cost-based architecture, the cost-based architecture including a Markov decision process (MDP) with a discounted horizon approach applied to pre-chosen open-loop optimistic policies that are time based, wherein the behavior planner uses the MDP for choosing among the pre-chosen open-loop optimistic policies based on respective costs associated with the pre-chosen open-loop optimistic policies, the costs determined by performing a rollout for at least one gap in a fixed time horizon; a motion planner receiving an output of the behavior planner based on the MDP; and a controller receiving an output of the motion planner and determining vehicle dynamics of the vehicle for a next timestep.
1 . An advanced driver-assistance system (ADAS) comprising:
a sensor;
a behavior planner that performs multi-policy lane change assistance for a vehicle by identifying multiple gaps for the multi-policy lane change assistance and evaluating multiple scenarios based on an output of the sensor using a cost-based architecture, the cost-based architecture including a Markov decision process (MDP) with a discounted horizon approach applied to pre-chosen open-loop optimistic policies that are time based, wherein the behavior planner uses the MDP for choosing among the pre-chosen open-loop optimistic policies by evaluating respective costs of each of the pre-chosen open-loop optimistic policies, the costs determined by performing a rollout for each of the gaps in a fixed time horizon such that if a lane where an actor is located can only be predicted for a certain length of time, then a behavior of the actor is predicted in the rollout only up until the certain length of time;
a motion planner receiving an output of the behavior planner based on the MDP; and
a controller receiving an output of the motion planner, determining vehicle dynamics of the vehicle for a next timestep, and controlling motion of the vehicle according to the vehicle dynamics of the vehicle for the next timestep.
2 . The ADAS of claim 1 , where the fixed time horizon is common to each of the gaps.
3 . The ADAS of claim 1 , wherein applying the discounted horizon approach comprises prioritizing events closer to the vehicle in time over events further away from the vehicle in time.
4 . The ADAS of claim 1 , wherein each of the pre-chosen open-loop optimistic policies comprises a combination of actions for the vehicle, including a fixed trajectory represented by a velocity profile.
5 . The ADAS of claim 1 , wherein the MDP comprises iterating for each of multiple gaps identified for the multi-policy lane change assistance.
6 . The ADAS of claim 5 , wherein the MDP further comprises iterating for each of the pre-chosen open-loop optimistic policies for each of the gaps.
7 . A method comprising:
receiving a sensor output from a sensor of an advanced driver-assistance system (ADAS);
performing multi-policy lane change assistance for a vehicle including:
identifying, by a processor, multiple gaps for the multi-policy lane change assistance:
applying, by the processor, in a Markov decision process (MDP), a discounted horizon approach to pre-chosen open-loop optimistic policies that are time based; and
evaluating, by the processor, multiple scenarios based on the sensor output using a cost-based architecture, including using the MDP for choosing among the pre-chosen open-loop optimistic policies by evaluating respective costs of each of the pre-chosen open-loop optimistic policies, the costs determined by performing a rollout for each of the gaps such that if a lane where an actor is located can only be predicted for a certain length of time;
generating, by the processor, a behavior planning output based on the MDP;
generating, by the processor, amotion planning output based on the behavior planning output;
determining, by the processor, vehicle dynamics of the vehicle for a next timestep based on the motion planning output; and
controlling, by the processor, motion of the vehicle according to the vehicle dynamics of the vehicle for the next timestep.
8 . The method of claim 7 , wherein applying the discounted horizon approach comprises prioritizing events closer to the vehicle.
9 . The method of claim 7 , wherein the MDP comprises iterating for each of multiple gaps identified for the multi-policy lane change assistance.
10 . The method of claim 7 , wherein the MDP comprises iterating for each of the pre-chosen open-loop optimistic policies for each of the gaps.