Systems and Methods for Optimized Multi-Agent Routing Between Nodes
Systems and methods described herein can provide for: obtaining, from a remote autonomous vehicle computing system, an incoming communication vector descriptive of a local environmental condition of a remote autonomous vehicle; inputting the incoming communication vector into a value iteration graph neural network of an autonomous vehicle; generating, by the value iteration graph neural network, transportation segment navigation instructions identifying a target transportation segment to navigate the autonomous vehicle to; generating a motion plan through an environment of the autonomous vehicle; and controlling the autonomous vehicle by one or more vehicle control systems based on the motion plan.
1 . A computer-implemented method comprising:
obtaining, from a remote autonomous vehicle computing system, an incoming communication vector descriptive of a local environmental condition of a remote autonomous vehicle;
inputting the incoming communication vector into a value iteration graph neural network of an autonomous vehicle;
generating, by the value iteration graph neural network, transportation segment navigation instructions identifying a target transportation segment to navigate the autonomous vehicle to;
generating a motion plan through an environment of the autonomous vehicle; and
controlling the autonomous vehicle by one or more vehicle control systems based on the motion plan.
2 . The computer-implemented method of claim 1 , further comprising generating an outgoing communication vector descriptive of one or more node values of the value iteration graph neural network.
3 . The computer-implemented method of claim 2 , wherein the outgoing communication vector further comprises information included in the incoming communication vector.
4 . The computer-implemented method of claim 2 , wherein the remote autonomous vehicle comprises a second value iteration graph neural network configured to receive the outgoing communication vector.
5 . The computer-implemented method of claim 1 , further comprising:
obtaining, from a plurality of additional remote autonomous vehicle computing systems, a plurality of additional incoming communication vectors respectively descriptive of respective local environmental conditions of a plurality of additional remote autonomous vehicles respective to the plurality of additional remote autonomous vehicle computing systems; and
inputting the plurality of additional incoming communication vectors into the value iteration graph neural network.
6 . The computer-implemented method of claim 5 , further comprising:
aggregating, by a machine-learned aggregation layer, the incoming communication vector and the plurality of additional incoming communication vectors to generate an aggregated incoming communication vector; and
inputting the aggregated incoming communication vector into the value iteration graph neural network.
7 . The computer-implemented method of claim 6 , wherein the machine-learned aggregation layer comprises attentional actor weights respectively associated with the remote autonomous vehicle and respectively associated with the plurality of additional remote autonomous vehicles.
8 . The computer-implemented method of claim 6 , wherein the machine-learned aggregation layer comprises an adjacency matrix.
9 . The computer-implemented method of claim 1 , wherein the value iteration graph neural network comprises a map graph, the map graph comprising one or more nodes corresponding to one or more transportation segments of a transportation segment network.
10 . The computer-implemented method of claim 9 , wherein the autonomous vehicle is positioned at a first segment of the one or more transportation segments of the transportation segment network, and wherein the remote autonomous vehicle is positioned at a second segment of the one or more transportation segments of the transportation segment network.
11 . The computer-implemented method of claim 1 , further comprising:
evaluating a value function of a plurality of nodes of the value iteration graph neural network; and
identifying a node of the plurality of nodes of the value iteration graph neural network having a maximum value of the value function respective to the node.
12 . The computer-implemented method of claim 1 , further comprising updating, based on the incoming communication vector, the value iteration graph neural network.
13 . The computer-implemented method of claim 12 , further comprising updating the value iteration graph neural network based on observational data of the autonomous vehicle.
14 . An autonomous vehicle (AV) computing system, comprising:
one or more processors; and
one or more tangible, non-transitory, computer-readable media that store instructions that are executable by the one or more processors to perform operations comprising:
obtaining, from a remote autonomous vehicle computing system, an incoming communication vector descriptive of a local environmental condition of a remote autonomous vehicle;
inputting the incoming communication vector into a value iteration graph neural network of an autonomous vehicle;
generating, by the value iteration graph neural network, transportation segment navigation instructions identifying a target transportation segment to navigate the autonomous vehicle to;
generating a motion plan through an environment of the autonomous vehicle; and
controlling the autonomous vehicle by one or more vehicle control systems based on the motion plan.
15 . The AV computing system of claim 14 , wherein the operations further comprise generating an outgoing communication vector descriptive of one or more node values of the value iteration graph neural network.
16 . The AV computing system of claim 14 , wherein the operations further comprise:
obtaining, from a plurality of additional remote autonomous vehicle computing systems, a plurality of additional incoming communication vectors respectively descriptive of respective local environmental conditions of a plurality of additional remote autonomous vehicles respective to the plurality of additional remote autonomous vehicle computing systems; and
inputting the plurality of additional incoming communication vectors into the value iteration graph neural network.
17 . The AV computing system of claim 16 , wherein the operations further comprise:
aggregating, by a machine-learned aggregation layer, the incoming communication vector and the plurality of additional incoming communication vectors to generate an aggregated incoming communication vector; and
inputting the aggregated incoming communication vector into the value iteration graph neural network.
18 . The AV computing system of claim 14 , wherein the value iteration graph neural network comprises a map graph, the map graph comprising one or more nodes corresponding to one or more transportation segments of a transportation segment network.
19 . The AV computing system of claim 14 , wherein the operations further comprise updating, based on the incoming communication vector, the value iteration graph neural network.
20 . An autonomous vehicle, comprising:
one or more processors; and
one or more tangible, non-transitory, computer-readable media that store instructions that are executable by the one or more processors to perform operations comprising:
obtaining, from a remote autonomous vehicle computing system, an incoming communication vector descriptive of a local environmental condition of a remote autonomous vehicle;
inputting the incoming communication vector into a graph neural network structured to update node feature vectors of a plurality of nodes that respectively correspond to a plurality of segments of a transportation network and to select, based on the updated node feature vectors, a target transportation segment of the plurality of segments;
generating, by the graph neural network, transportation segment navigation instructions identifying the target transportation segment to navigate the autonomous vehicle to;
generating a motion plan through an environment of the autonomous vehicle; and
controlling the autonomous vehicle by one or more vehicle control systems based on the motion plan.