Huawei Technologies Co., Limited

Other

Huawei Technologies Canada Co., Ltd.

Peyman Yadmellat

Yunfei Zhang

Seyed Masoud Nosrati

Mohsen Rohani

Elmira Amirloo Abolfathi

Kasra Rezaee

Hao Chen

Hengshuai Yao

Jun Luo

Barry J Gilhuly

Stephen L Smith

Mohammed Elmahgiubi

Armin Sadeghi

Han Hu

Weize Zhang

Systems, methods and computer-readable media for spatio-temporal motion planning, including: receiving data defining a drivable area within a spatio-temporal space; generating a cost function to be minimized that includes one or more upper-bound cost terms, each upper-bound cost term approximating the upper-bound of a cost term for the trajectory for a spatial frame of the spatio-temporal space, each upper-bound cost term being expressed as one or more longitudinal-temporal cost terms and one or more lateral-temporal cost terms for the trajectory; and further cost terms for the trajectory for the spatio-temporal frames; computing, based on the constraints and cost function, a planned trajectory through the drivable area, the planned trajectory directory being optimized with respect to both spatial frame and the spatio-temporal frames. Optimizing the planned trajectory in the spatial frame in addition to the spatio-temporal frames can enable planning objectives for the spatial frame to be optimized, for example maintaining closeness to a reference path and minimizing lateral acceleration and directional oscillation.

Systems, methods, and computer-readable media for spatio-temporal motion planning

Methods and systems of training RL agent for autonomous operation of a vehicle are described. The RL agent is trained using uniformly sampled training samples and learning a policy. After the RL agent has achieved a predetermined performance goal, data is collected including a sequence of sampled states, and for each sequence of sampled states, agent parameters, and an indication of failure of the RL agent for the sequence. A failure predictor is trained, using samples from the collected data, to predict a probability of failure of the RL agent for a given sequence of states. Sequences of states are collected by simulating interaction of the vehicle with the environment. Based on a probability of failure outputted by the failure predictor, a sequence of states is selected. The RL agent is further trained based on the selected sequence of states.

Method and system for training reinforcement learning agent using adversarial sampling

The present disclosure relates to methods and systems for generating a channel for use as a spatio-temporal constraint for motion planning for an autonomous vehicle. The method generates a triangulation mesh for a space in an environment. The triangulation mesh includes a plurality of nodes, each represents a geographic location in the space. Some of the nodes correspond to dynamic objects. Based on the triangulation mesh, a candidate channel extending from a starting location to a target location of the autonomous vehicle is generated. Further, the method predicts a time and location of a future triangulation mesh topology event caused by at least one of the nodes that will impact the candidate channel. A valid channel segment for the channel is selected, which extends from the starting location to a channel segment end location preceding the predicted location of the future triangulation mesh topology event. The channel improves the validity of motion planning for a longer time period.

Method and system for motion planning for an autonmous vehicle

Methods and systems for training a motion planner for an autonomous vehicle are described. A trajectory evaluator agent of the motion planner receives state data defining a current state of the autonomous vehicle and an environment at a current time step. Based on the current state, a trajectory is selected. A reward is calculated based on performance of the selected trajectory in the current state. State data is received for a next state of the autonomous vehicle and the environment at a next time step. Parameters of the trajectory evaluator agent are updated based on the current state, selected trajectory, computed reward and next state. The parameters of the trajectory evaluator agent are updated to assign an evaluation value for the selected trajectory that reflects the calculated reward and expected performance of the selected trajectory in the future states.

Predictive motion planning system and method

Systems, methods and computer-readable media for selecting a trajectory for an autonomous vehicle are disclosed that include computing a current vehicle state for the autonomous vehicle based on observations by a sensing system; computing respective collision probability scores for a plurality of candidate trajectories based on the current vehicle state; computing respective information gain scores for the plurality of candidate trajectories based on the current vehicle state, the information gain score for each candidate trajectory indicating an respective information gain for a next planning horizon interval that is subsequent to the current planning horizon interval; and selecting a planned trajectory from the plurality of candidate trajectories based on the respective collision probability scores and respective information gain scores.

Systems, methods, and media for occlusion-aware motion planning

A processor-implemented method and system for determining a predictive occupancy grid map (OGM) for an autonomous vehicle are disclosed. The method includes: receiving a set of OGMs including a current predicted OGM and one or more future predicted OGMs, the current OGM associated with a current timestamp and each future predicted OGM associated with a future timestamp; generating a weight map associated with the current timestamp based on one or more kinodynamic parameters of the vehicle at the current time stamp, and one or more weight map associated with a future timestamp; generating a set of filtered predicted OGMs by filtering the current predicted OGM with the weight map associated the current timestamp and filtering each respective future predicted OGM associated with a future timestamp with the weight map associated with the respective future timestamp; and sending a single predicted OGM to a trajectory generator.

System and method for predictive path planning in autonomous vehicles

A method and system for determining a trajectory within an operating space for an autonomous vehicle to implement a behaviour decision, comprising: generating a set of candidate target end states for the behaviour decision based on an estimated state of the vehicle; generating a set of candidate trajectories corresponding to the set of candidate target end states based on the estimated state of the vehicle; determining a suitability of each of the candidate target end states based on the estimated state of the vehicle; and selecting a trajectory to implement the behaviour decision from the set of candidate trajectories based on the determined suitability of the candidate target end states.

System and method for hierarchical planning in autonomous vehicles

A method, device and system of prediction of a state of an object in the environment using an action model of a neural network. In accordance with one aspect, a control system for a object comprises a processor, a plurality of sensors coupled to the processor for sensing a current state of the object and an environment in which the object is located, and a first neural network coupled to the processor. One or more predicted subsequent states of the object in the environment are determined using an action model of the neural network and a current state of the object in the environment and an plurality of action sequences. The action model comprises a mapping of states of the object in the environment and actions performed by the object for each state to predicted subsequent states of the object in the environment.

Method of prediction of a state of an object in the environment using an action model of a neural network

A method, device and system of prediction of a state of an object in the environment using a pre-trained action model defined by an action model neural network. A control system for an object comprises a plurality of sensors for sensing a current state and an environment in which the object is located, and a first neural network. Predicted subsequent states of the object in the environment are obtained using the action model and a current state of the object in the environment The action model maps a plurality of state-action pairs (s, a), each state-action pair encoding a state (s) of the object in the environment and an action (a) performed by the object to a predicted subsequent state (s') of the object in the environment. An action that maximizes a value of a target, based at least on a reward for each of the predicted subsequent states, is determined. The determined action is caused to be performed.

Method of selection of an action for an object using a neural network

Methods and systems for controlling motion of a vehicle. Sensor data is obtained, representing an environment of the vehicle. A reference control trajectory is corrected, using one or more virtual forces, to provide a desired trajectory. The one or more virtual forces are generated by applying an impedance scheme to the sensor data. A desired trajectory is outputted for controlling the vehicle.

Growing community of inventors

North York, Canada

Peyman Yadmellat