Abstract: Techniques are discussed herein for generating trajectories for controlling motion and/or other behaviors of vehicles in driving environments. In particular, techniques are described herein for using a tree search to determine a trajectory for a vehicle to join a driving route from an initial vehicle state off of the driving route structure. A vehicle computing system may determine various candidate trajectories, including trajectories based on an off-route inertial reference frame, additional trajectories based on the route structure, perturbed trajectories, etc. The set of candidate trajectories may be optimized and/or filtered based on objects in the environment, and the corresponding candidate actions may be used to generate a search tree between the off-route vehicle state and an on-route target state.

Abstract: Techniques for procedurally generating virtual representations of buildings within a simulated environment using building segments from artist-rendered facades are described. The facades are input into the system and the techniques provide for dividing the facade into building segments based on features of the facade. The building segments are assigned rules for assembling the building sections together to populate a virtual building. The rules are manipulable to change possible locations and orientations for the building sections. The building sections are procedurally placed within a building frame according to the rules and the virtual building placed within a simulated driving environment for testing one or more vehicle systems.

Abstract: According to an aspect of the invention, there is provided a system comprising: a processor; and non-transitory memory storing processor-executable instructions that, when executed by the processor, cause the system to perform actions including: receiving a driving log comprising data relating to a real object in a real environment; controlling, in a driving simulation generated based on the driving log, a simulated object corresponding to the real object according to an object model; determining a simulated distribution for a characteristic that is controlled according to the object model, the simulated distribution being based on a value for the characteristic associated with the simulated object based on the driving simulation; determining a real distribution for the characteristic, the real distribution being based on a value for the characteristic associated with the real object based on the driving log; and determining a metric based on the simulated distribution and the real distribution.

Abstract: A hierarchical tree search may determine different costs associated with a same candidate action for different levels of the hierarchy, each of which may be associated with different cost function(s) and respective objective(s), such as safety, progress, comfort, and the like. At each level, the hierarchical tree search may determine an upper and lower bound cost for the candidate action that may be based on the cost function(s) for that level. The hierarchical tree search may mask out for subsequent level(s) any candidate actions at a level of the tree that exceed the lowest upper or lower bound cost of any candidate action at that level by more than a slack amount.

Abstract: Techniques are provided for optimizing the performance of each of the components of a vehicle thermal system including a radiator system and an air conditioning system to provide desired cooling inside the vehicle cabin while simultaneously reducing the noise and potential vibration caused by operation of the individual components of the thermal system and by the operation of the components collectively. Optimizing the operating settings of each component relative to each other provides for desired interior vehicle cabin cooling while maintaining efficient thermal loading and energy use because while the speed and/or duty cycle of one component may be increased, the speed and/or duty cycle of another component may be decreased resulting in better overall thermal loading and energy efficiency.

Abstract: Techniques for optimal trajectory generation for a vehicle are described herein. In some cases, the techniques described herein include determining a first primary trajectory by a first component (e.g., a planner component) of the vehicle, determining a triggering event associated with the first primary trajectory by a second component (e.g., a trajectory validation component) of the vehicle, controlling the vehicle based on an alternative trajectory (e.g., a trajectory configured to cause the vehicle to stop and/or slow down) in response to the triggering event, determining a second primary trajectory by the first component and based at least in part on a state of the vehicle along the alternative trajectory, and one or more of: (i) controlling the vehicle based on the alternative trajectory if the triggering event is still present, or (ii) controlling the vehicle based on the second primary trajectory if the triggering event is no longer present.

Abstract: There is provided a system configured to receive data associated with a vehicle operating within an environment; generate, based at least in part on the data, a graph comprising a plurality of nodes, a node of the plurality associated with one or more of a vehicle operating in the environment, a road feature, an additional vehicle, or a pedestrian; input the graph into a self-supervised machine learned model comprising an encoder, wherein the machine learned model is trained to output a representation associated with the node; receive, from the self-supervised machine learned model, a representation associated with the node; and transmit the representation to a downstream machine learned model trained to output control data based at least in part on the representation, wherein the control data is configured to control the vehicle or another vehicle.

Abstract: Techniques are discussed for vehicle fault management, including determining and applying time-dependent and/or distance-dependent constraints to enable the vehicle to respond safely and appropriately to faults. In various examples, a fault may be detected by a vehicle component or system, indicating a condition necessitating the vehicle to cease operating on the road surface within a time- or distance-based driving constraint. Additional techniques may include determining, via a route planning component and/or a remote control system, a modified destination such as a pullover location for the vehicle. The modified destination may be determined based on the time- and/or distance-based constraint determined based on one or more vehicle faults, map data of the driving environment, and/or factors such as the driving conditions and the state of additional vehicles in the vehicle fleet.

Abstract: Techniques for determining whether a global navigation satellite system (GNSS) measurement generated by a sensor associated with a vehicle is reliable. A system determines whether a first GNSS measurement generated by a first sensor and associated with a first time is reliable based on at least one of: (i) one or more GNSS measurements generated by the first sensor and associated with one or more times before and/or after the first time, or (ii) one or more GNSS measurements generated by a second sensor and associated with the first time. The system may determine whether a first GNSS measurement generated by a first sensor and associated with a first time is reliable based on whether a ratio of a sum of incremental distances associated with a GNSS measurement sequence including the first GNSS measurement over an odometry-based distance falls within a threshold range.

Abstract: A machine-learned architecture for estimating the cost to operate from a prediction node of a tree search for exploring candidate operations for controlling a vehicle may be trained using a loss function that mitigates the deleterious effects of training without infinitely searching the tree or reaching convergence. A first loss of the loss function term may be based on a difference between the architecture's estimated cost to go from a starting position and the cost to go from the starting position determined by the search so far. A second loss term may be based on the difference between the model's estimate of the cost to go from a predicted state and may be adjusted based on (1) weighting based on a convergence-based weight and/or (2) adjusting the second loss term using an approximation of a difference between a cost of the current cheapest action and the globally optimal action.

Abstract: Techniques are described herein for encoding driving scenarios based on the characterizations of interactions between pairs or groups of agents within the driving scenarios. Additional techniques are described herein for executing driving simulations and using machine-learned (ML) classifier models to triage collisions and other interactions within the driving simulations. Further techniques are described for controlling agents and/or managing agent interactions within such driving simulations in a more realistic manner, including determining agent right-of-way and/or resolving headlock negotiations between programmatic agents in the simulations.

Abstract: Various embodiments relate generally to autonomous vehicles and associated mechanical, electrical and electronic hardware, computer software and systems, and wired and wireless network communications to provide an autonomous vehicle fleet as a service. In particular, a method may include receiving first sensor data from a first sensor disposed on a vehicle, the first sensor data associated with a first sensor modality, and receiving second sensor data from a second sensor disposed on the vehicle, the second sensor data associated with a second sensor modality different than the first sensor modality. The method may further include generating fused sensor data representing at least a portion of the first sensor data and the second sensor data, generating a trajectory for the vehicle based in part on the fused sensor data, and controlling the vehicle based in part on the trajectory.

Abstract: Techniques for identifying false-positives caused by sensor deflection off of reflective surfaces. The techniques may include receiving data indicating a presence of an object in an environment of a vehicle. Based at least in part on the data, a determination may be made that the object is occluded from view by another object in the environment. The techniques may also include determining that a retroreflective surface is located within a threshold proximity of the other object in the environment. Based at least in part on the retroreflective surface being located within the threshold proximity of the other object, a determination may be made that the presence of the object is caused by a false-positive condition associated with a sensor of the vehicle reflecting off of the retroreflective surface. In some instances, the object may be labeled as a false-positive object that is absent from the environment.

Abstract: Techniques for storing predicted state using a linear array and for querying the linear array to evaluate a proposed trajectory are described herein. In some cases, a first processing unit (e.g., a central processing unit (CPU)) determines prediction data associated with an object in the vehicle environment and sends the prediction data to a second processing unit (e.g., a graphics processing unit (GPU)). In some cases, after the second processing unit receives the prediction data, the second processing unit stores the prediction data using a linear array. In some cases, the second processing unit can use the linear array to evaluate a proposed trajectory for the vehicle and determine whether to validate the proposed trajectory for the vehicle. In some cases, based on the validation of the proposed trajectory by the second processing unit, the vehicle is controlled based on the proposed trajectory.

Abstract: A side-impact crash structure may be positioned proximate an end of a vehicle to reduce impact forces imparted to an occupant during a side collision. The side-impact crash structure may include energy absorbers positioned on lateral sides of the vehicle between the passenger compartment and a longitudinal end of the vehicle. The energy absorbers may be configured to absorb impact forces over a limited ride-down distance (the distance over which the deceleration occurs) to prevent damage to both the occupants and one or more components or systems of the vehicle. The energy absorbers can be configured to deform along an oblique axis of the vehicle under a compressive force. The side-impact crash structure may include one or more load spreaders disposed between the energy absorbers that disperse the impact force to other portions of the vehicle.

Abstract: Techniques for determining labels and occupancy data for voxels and pixels representing object protrusions are disclosed. The occupancy status of voxels surrounding an occupied voxel is determined and used to determine the occupancy density of the occupied voxel. The loss for the occupied voxel is adjusted inversely proportionately to the occupancy density. The adjusted-loss voxel is used to train a machine-learned model to detect objects in an environment and, specifically, to more accurately detect objects having protrusions that may otherwise not be associated with the object. This model may be used to provide data used to control a vehicle.

Abstract: Techniques for determining behavior profiles and unique identities for objects observed in an environment surrounding an autonomous vehicle are described. The behavior profiles may be generated and associated with unique identifiers that are stored and used by a fleet of vehicles for planning and navigating in traffic environments. The behavior profiles may be used to control operation of the vehicle and increase safety of the vehicle by identifying potentially risky or erratic objects and navigating through the environment accordingly. The behavior profiles may be stored at a central database and propagated across a fleet to increase observations used to build profiles and thereby increase profile confidence.

Abstract: Techniques for estimating a ground profile, including a ground height (e.g., elevation) and ground roll, for an environment associated with a vehicle are described herein. Sensor data associated with a vehicle may be used to determine sensor data points associated with a path of the vehicle. Based on an upper bound slope and lower bound slope, such as a maximum and/or minimum rate of ground roll change, a region (e.g., search window) may be determined. Based on the region, a subset of sensor data points included in the region may be determined. The subset of sensor data points may be used to determine a representative height and representative roll. In some instances, the representative height and representative roll may be determined using a Hough transform and the subset of sensor data points. A ground profile may be determined based on the representative height and representative roll.