Object determination in an occluded region

One or more non-transitory computer-readable storage media storing instructions that, when executed, cause one or more processors to perform operations comprising:

• receiving, first occlusion data associated with a first occluded region in an environment of a vehicle, wherein the first occlusion data is associated with a first time and a first weight determined based at least in part on the first time;
• receiving, second occlusion data associated with the first occluded region in the environment, the second occlusion data associated with a second time after the first time, wherein the second occlusion data is associated with a second weight determined based at least in part on the second time, wherein the second weight is higher than the first weight;
• fusing, based at least in part on the first weight and the second weight, the first occlusion data and the second occlusion data, to generate fused data of the first occluded region and excluding other portions of the environment;
• determining, based at least in part on the fused data, a first probability of an object occupying the first occluded region;
• determining, based at least in part on the first occlusion data and the second occlusion data, a second probability that a second occluded region of the environment is associated with an additional object;
• receiving a candidate trajectory for the vehicle to follow;
• determining, based at least in part on the first probability, a distance between a closest point of the first occluded region and a point of the vehicle along the candidate trajectory;
• identifying, based at least in part on the first probability and the second probability, one of the object of the first occluded region or the additional object of the second occluded region as having a greater likelihood of intersecting with the vehicle relative to the object of the first occluded region or the additional object of the second occluded region;
• and
• controlling operation of the vehicle based at least in part on the distance, wherein controlling the operation of the vehicle is further based at least in part on the identifying.

A method comprising:

• inputting, into a model, first region data associated with an occluded region at a first time;
• inputting, into the model, second region data associated with the occluded region, wherein the second region data is associated with a second time different from the first time;
• inputting, into the model, a trajectory of a vehicle traversing an environment;
• fusing, as a fused representation of the occluded region and excluding other regions of the environment, the first region data and the second region data;
• receiving, from the model, a plurality of probabilities in a discretized grid, wherein individual ones of the plurality of probabilities represent a discrete probability that a potential object within the occluded region intersects with the vehicle at the second time, wherein the potential object is associated with an object class, and the discrete probability that the potential object within the occluded region intersects with the vehicle at the second time is based at least in part on the fused representation;
• determining a potential adversarial behavior of the potential object, wherein the potential adversarial behavior of the potential object is determined based on one or more characteristics associated with the object class, and the one or more characteristics comprises one or more of a maximum velocity associated with the object class, a maximum acceleration associated with the object class, or a maximum turning ability associated with the object class;
• and
• controlling operation of the vehicle at the second time based at least in part on the plurality of probabilities and the potential adversarial behavior of the potential object.