Systems, methods, tangible non-transitory computer-readable media, and devices associated with the motion prediction and operation of a device including a vehicle are provided. For example, a vehicle computing system can access state data including information associated with locations and characteristics of objects over a plurality of time intervals. Trajectories of the objects at subsequent time intervals following the plurality of time intervals can be determined based on the state data and a machine-learned tracking and kinematics model. The trajectories of the objects can include predicted locations of the objects at subsequent time intervals that follow the plurality of time intervals. Further, the predicted locations of the objects can be based on physical constraints of the objects. Furthermore, indications, which can include visual indications, can be generated based on the predicted locations of the objects at the subsequent time intervals.

A motion planner performs motion planning with collision assessment, using a motion planning lattice that represents configuration states of a primary agent (e.g., autonomous vehicle) as nodes and transitions between states as edges. The system may assign cost values to edges, the cost values representing probability or likelihood of collision for the corresponding transition. The cost values may additionally or alternatively represent a severity of collision, for example generated via a parametric function with two or more parameters and one or more weights. A primary agent and/or dynamic obstacles may be represented as respective oriented bounding boxes. Some obstacles (e.g., road markings, edge of road) may be represented as curves. A trajectory of a primary agent and/or dynamic obstacle may be represented by respective sets of fitted polynomial functions, edges on the planning graph, which represent transitions in states of the primary agent, the system sets value representing a probability of collision, and optionally representing a severity of the collision. The system then causes the actuator system of the primary agent to implement a motion plan with the applicable identified path based at least in part on the optimization.

Systems and methods for motion planning by a vehicle computing system of an autonomous vehicle are provided. The vehicle computing system can input sensor data to a machine-learned system including one or more machine-learned models. The computing system can obtain, as an output of the machine-learned model(s), motion prediction(s) associated with object(s) detected by the system. The system can convert a shape of the object(s) into a probability of occupancy by convolving an occupied area of the object(s) with a continuous uncertainty associated with the object(s). The system can determine a probability of future occupancy of a plurality of locations in the environment at future times based at least in part on the motion prediction(s) and the probability of occupancy of the object(s). The system can provide the motion prediction(s) and the probability of future occupancy of the plurality of locations to a motion planning system of the autonomous vehicle.

A graph representation of a tactical map representing a plurality of static components of an environment of a vehicle is generated. Nodes of the graph represent static components, and edges represent relationships between multiple static components. Different edge types are used to indicate respective relationship semantics among the static components. Individual nodes are represented as having the same number and types of edges in the graph. Using the graph as input to a neural network based model, a set of results is obtained. A motion control directive based at least in part on the results is transmitted to a motion-control subsystem of the vehicle.

A method comprises receiving driving data from a plurality of connected vehicles approaching an intersection, the driving data comprising a speed and position of a connected vehicle, determining estimated times of arrival that each of the connected vehicles will arrive at the intersection based on the driving data, scheduling the connected vehicles to cross the intersection in a particular order based on the estimated times of arrival, and transmitting the scheduled order to the connected vehicles.

Among other things, techniques are described for predicting how an agent (e.g., a vehicle, bicycle, pedestrian, etc.) will move in an environment based on prior movement, the road network, the surrounding objects and/or other relevant environmental factors. One trajectory prediction technique involves generating a probability map for an agent's movement. Another trajectory prediction technique involves generating a trajectory lattice, for an agent's movement. In addition, a different trajectory prediction technique involves multi-modal regression where a classifier (e.g., a neural network) is trained to classify the probability of a number of (learned) modes such that each model produces a trajectory based on the current input.

A first possible space, for which a lane change of the vehicle is possible, is specified from an inter-vehicle distance between the preceding other vehicle and the nearby vehicle, a speed of the preceding other vehicle, and a speed of the nearby vehicle. A second possible space, for which the lane change of the vehicle is possible, is specified from an inter-vehicle distance between the nearby vehicle and the following other vehicle, the speed of the nearby vehicle, and a speed of the following other vehicle. The control unit controls the traveling of the vehicle to make the lane change to the adjacent lane based on a result of the evaluation of the first possible space and the second possible space.

Methods and systems for out-of-distribution (OOD) detection in autonomous driving systems are described. A method for use in an autonomous driving system may include filtering feature vectors. The feature vectors may be filtered using a first filter to obtain clusters of feature vectors. The method may include assigning one or more images to a respective cluster based on a feature vector of the image. The method may include filtering a subset of the images using a second filter to determine a classification model. The method may include storing the classification model on a vehicle control system of a vehicle. The method may include detecting an image using a vehicle sensor. The method may include classifying the detected image based on the classification model. The method may include performing a vehicle action based on the classified detected image.

A system for generating a model-free reinforcement learning policy may include a processor, a memory, and a simulator. The simulator may be implemented via the processor and the memory. The simulator may generate a simulated traffic scenario including two or more lanes, an ego-vehicle, a dead end position, and one or more traffic participants. The dead end position may be a position by which a lane change for the ego-vehicle may be desired. The simulated traffic scenario may be associated with an occupancy map, a relative velocity map, a relative displacement map, and a relative heading map at each time step within the simulated traffic scenario. The simulator may model the ego-vehicle and one or more of the traffic participants using a kinematic bicycle model. The simulator may build a policy based on the simulated traffic scenario using an actor-critic network. The policy may be implemented on an autonomous vehicle.

The disclosure provides a method for operating an autonomous vehicle. To operate the autonomous vehicle, a plurality of lane segments that are in an environment of the autonomous vehicle is determined and a first object and a second object in the environment are detected. A first position for the first object is determined in relation to the plurality of lane segments, and particular lane segments that are occluded by the first object are determined using the first position. According to the occluded lane segments, a reaction time is determined for the second object and a driving instruction for the autonomous vehicle is determined according to the reaction time. The autonomous vehicle is then operated based on the driving instruction.