Machine learning enhanced vehicle merging
A method receives a first image set depicting a merging zone, the first image set including first image(s) associated with a first timestamp; determines, using a trained first machine learning logic, a first state describing a traffic condition of the merging zone at the first timestamp using the first image set; determines, from a sequence of states describing the traffic condition of the merging zone at a sequence of timestamps, using a trained second machine learning logic, second state(s) associated with second timestamp(s) prior to the first timestamp of the first state using a trained backward time distance; computes, using a trained third machine learning logic, impact metric(s) for merging action(s) using the first state, the second state(s), and the merging action(s); selects, from the merging action(s), a first merging action based on the impact metric(s); and provides a merging instruction including the first merging action to a merging vehicle.
1. A method comprising:
receiving a first image set depicting a merging zone, the first image set including one or more first images associated with a first timestamp;
determining, using a trained first machine learning logic, a first state describing a traffic condition of the merging zone at the first timestamp using the first image set;
determining, from a sequence of states describing the traffic condition of the merging zone at a sequence of timestamps, using a trained second machine learning logic, one or more second states associated with one or more second timestamps prior to the first timestamp of the first state using a trained backward time distance;
computing, using a trained third machine learning logic, one or more impact metrics for one or more merging actions using the first state, the one or more second states, and the one or more merging actions;
selecting, from the one or more merging actions, a first merging action based on the one or more impact metrics; and
providing a merging instruction including the first merging action to a merging vehicle, the merging instruction instructing the merging vehicle to perform the first merging action in the merging zone.
2. The method of claim 1 , wherein the one or more second states includes a plurality of second states, and determining the one or more second states from the sequence of states includes:
determining a past time range based on the first timestamp of the first state and the trained backward time distance; and
determining, from the sequence of states, the plurality of second states associated with a plurality of second timestamps within the past time range.
3. The method of claim 1 , wherein the one or more merging actions include one or more speed adjustments to one or more target merging speeds, one or more acceleration adjustments to one or more target acceleration rates, and one or more angle adjustments to one or more target steering angles.
4. The method of claim 1 , further comprising:
processing, using a control unit of the merging vehicle, the merging instruction including the first merging action; and
controlling, using the control unit of the merging vehicle, the merging vehicle to perform the first merging action in the merging zone.
5. The method of claim 1 , wherein
the merging zone includes an on-ramp segment, a upstream freeway segment, and a downstream freeway segment, and
the first image set includes the one or more first images associated with the first timestamp received from one or more road side units located in the merging zone, one or more merging vehicles located on the on-ramp segment, one or more upstream mainline vehicles located on the upstream freeway segment, and one or more downstream mainline vehicles located on the downstream freeway segment.
6. The method of claim 1 , wherein the merging zone includes an on-ramp segment, a upstream freeway segment, and a downstream freeway segment, the method includes:
determining, using the trained first machine learning logic, each state in the sequence of states associated with a corresponding timestamp in the sequence of timestamps using an image set associated with the corresponding timestamp; and wherein
the image set includes one or more images associated with the corresponding timestamp received from one or more road side units located in the merging zone, one or more merging vehicles located on the on-ramp segment, one or more upstream mainline vehicles located on the upstream freeway segment, and one or more downstream mainline vehicles located on the downstream freeway segment.
7. The method of claim 1 , wherein each state in the sequence of states and the first state includes
roadway data describing one or more roadway components of the merging zone,
first vehicle movement data of one or more merging vehicles, second vehicle movement data of one or more upstream mainline vehicles, and third vehicle movement data of one or more downstream mainline vehicles, and
first segment traffic data of an on-ramp segment, second segment traffic data of a upstream freeway segment, and third segment traffic data of a downstream freeway segment.
8. The method of claim 7 , wherein
the vehicle movement data of each vehicle included in the one or more merging vehicles, the one or more upstream mainline vehicles, and the one or more downstream mainline vehicles specifies one or more of a vehicle location, a vehicle speed, and an acceleration pattern associated with the vehicle, and
the segment traffic data of each road segment among the on-ramp segment, the upstream freeway segment, and the downstream freeway segment specifies one or more of a traffic flow, a vehicle density, an average vehicle speed, and an average following distance associated with the road segment.
9. A system comprising:
one or more processors;
one or more memories storing instructions that, when executed by the one or more processors, cause the system to:
receive a first image set depicting a merging zone, the first image set including one or more first images associated with a first timestamp;
determine, using a trained first machine learning logic, a first state describing a traffic condition of the merging zone at the first timestamp using the first image set;
determine, from a sequence of states describing the traffic condition of the merging zone at a sequence of timestamps, using a trained second machine learning logic, one or more second states associated with one or more second timestamps prior to the first timestamp of the first state using a trained backward time distance;
compute, using a trained third machine learning logic, one or more impact metrics for one or more merging actions using the first state, the one or more second states, and the one or more merging actions;
select, from the one or more merging actions, a first merging action based on the one or more impact metrics; and
provide a merging instruction including the first merging action to a merging vehicle, the merging instruction instructing the merging vehicle to perform the first merging action in the merging zone.
10. The system of claim 9 , wherein
the one or more second states includes a plurality of second states, and
to determine the one or more second states from the sequence of states includes:
determining a past time range based on the first timestamp of the first state and the trained backward time distance; and
determining, from the sequence of states, the plurality of second states associated with a plurality of second timestamps within the past time range.
11. The system of claim 9 , wherein the one or more merging actions include one or more speed adjustments to one or more target merging speeds, one or more acceleration adjustments to one or more target acceleration rates, and one or more angle adjustments to one or more target steering angles.
12. The system of claim 9 , wherein the instructions, when executed by the one or more processors, further cause the system to:
process, using a control unit of the merging vehicle, the merging instruction including the first merging action; and
control, using the control unit of the merging vehicle, the merging vehicle to perform the first merging action in the merging zone.
13. The system of claim 9 , wherein
the merging zone includes an on-ramp segment, a upstream freeway segment, and a downstream freeway segment, and
the first image set includes the one or more first images associated with the first timestamp received from one or more road side units located in the merging zone, one or more merging vehicles located on the on-ramp segment, one or more upstream mainline vehicles located on the upstream freeway segment, and one or more downstream mainline vehicles located on the downstream freeway segment.
14. The system of claim 9 , wherein
the merging zone includes an on-ramp segment, a upstream freeway segment, and a downstream freeway segment,
the instructions, when executed by the one or more processors, further cause the system to determine, using the trained first machine learning logic, each state in the sequence of states associated with a corresponding timestamp in the sequence of timestamps using an image set associated with the corresponding timestamp, and
the image set includes one or more images associated with the corresponding timestamp received from one or more road side units located in the merging zone, one or more merging vehicles located on the on-ramp segment, one or more upstream mainline vehicles located on the upstream freeway segment, and one or more downstream mainline vehicles located on the downstream freeway segment.
15. The system of claim 9 , wherein each state in the sequence of states and the first state includes
roadway data describing one or more roadway components of the merging zone,
first vehicle movement data of one or more merging vehicles, second vehicle movement data of one or more upstream mainline vehicles, and third vehicle movement data of one or more downstream mainline vehicles, and
first segment traffic data of an on-ramp segment, second segment traffic data of a upstream freeway segment, and third segment traffic data of a downstream freeway segment.
16. The system of claim 15 , wherein
the vehicle movement data of each vehicle included in the one or more merging vehicles, the one or more upstream mainline vehicles, and the one or more downstream mainline vehicles specifies one or more of a vehicle location, a vehicle speed, and an acceleration pattern associated with the vehicle, and
the segment traffic data of each road segment among the on-ramp segment, the upstream freeway segment, and the downstream freeway segment specifies one or more of a traffic flow, a vehicle density, an average vehicle speed, and an average following distance associated with the road segment.