Huawei Technologies Co., Limited

The Royal Institution for the Advancement of Learning / Mcgill University

Huawei Technologies Canada Co., Ltd.

Elmira Amirloo Abolfathi

Jun Luo

Mohsen Rohani

Peyman Yadmellat

Jason Philip Ku

Gregory Lewis Dudek

Yunfei Zhang

Keyvan Golestan Irani

Seyed Masoud Nosrati

David Meger

Lucas Pagé-Caccia

Joelle Pineau

Florian Shkurti

Yasasa Abeysirigoonawardena

Juan Camilo Gamboa Higuera

Mohammed Elmahgiubi

Melissa Mozifian

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

A system and method for path and/or motion planning and for training such a system are described. In one aspect, the method comprises generating a sequence of predicted occupancy grid maps (OGMs) for T&#x2212;Ttime steps based on a sequence of OGMs for 0&#x2212;Ttime steps, a reference map of an environment in which an autonomous vehicle is operating, and a trajectory. A cost volume is generated for the sequence of predicted OGMs. The cost volume comprises a plurality of cost maps for T&#x2212;Ttime steps. Each cost map corresponds to a predicted OGM in the sequence of predicted OGMs and has the same dimensions as the corresponding predicted OGM. Each cost map comprises a plurality of cells. Each cell in the cost map represents a cost of the cell in corresponding predicted OGM being occupied in accordance with a policy defined by a policy function.

Machine-learning based system for path and/or motion planning and method of training the same

Method or system for reinforcement learning that simultaneously learns a DR distribution &#x3d5; while optimizing an agent policy &#x3a0; to maximize performance over the learned DR distribution; method or system for training a learning agent using data synthesized by a simulator based on both a performance of the learning agent and a range of parameters present in the synthesized data.

Learning domain randomization distributions for transfer learning

A system, device and method of generating high resolution and high accuracy point cloud. In one aspect, a computer vision system receives a camera point cloud from a camera system and a LiDAR point cloud from a LiDAR system. An error of the camera point cloud is determined using the LiDAR point cloud as a reference. A correction function is determined based on the determined error. A corrected point cloud is generated from the camera point cloud using the correction function. A training error of the corrected point cloud is determined using the first LiDAR point cloud as a reference. The correction function is updated based on the determined training error. When training is completed, the correction function can be used by the computer vision system to generate a generating high resolution and high accuracy point cloud from the camera point cloud provided by the camera system.

System, device and method of generating a high resolution and high accuracy point cloud

A method for generation of an augmented point cloud with point features from aggregated 3D coordinate data and related device. The method comprises receiving a current point cloud in the form of 3D coordinate data in ego coordinates from one or more detection and ranging (DAR) devices of a vehicle. Features are extracted from the current point cloud. A previous point cloud is transformed into ego coordinates using a current location of the vehicle. Each point in the previous point cloud is transformed to align with a corresponding point in the current point cloud to generate a transformed point cloud. The current point cloud is aggregated with the transformed point cloud to generate an aggregated point cloud. The current point features are aggregated with the point features of the transformed point cloud to generate aggregated point features.

Method for generation of an augmented point cloud with point features from aggregated temporal 3D coordinate data, and related device

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

Methods and systems for generating synthetic point cloud data are described. Projected 2D data grid is generated by projecting a 3D point cloud into a 2D grid, with rotation equivariance. A generative model is learned using the projected 2D data grid, wherein the generative model is implemented using flex-convolution and transpose flex convolution operations, for example in a generative adversarial network. The learned generative model is used to generate synthetic point clouds.

Method and system for generating synthetic point cloud data using a generative model

A method and apparatus for generating adversarial scenarios and training an autonomous driving agent for an autonomous vehicle, using one or more sets of parameters, each set of parameters defining a respective driving scenario. A new set of parameters is generated by changing one or more parameters of one of the sets of parameters to define a new driving scenario, and performance of the autonomous driving agent is evaluated on the new driving scenario. The generating and evaluating is repeated until the autonomous driving agent fails to satisfy a predefined performance threshold for the new driving scenario. Each instance of changing the one or more parameters is based on a prior evaluated performance of the autonomous driving agent. The autonomous driving agent is trained to update a learned policy of the autonomous driving agent using at least one set of parameters, including the new set of parameters.

Iterative generation of adversarial scenarios

A system, method, and processor-readable medium for assessing the reliability of vehicle systems used in an autonomous vehicle. The assessment may be performed at least in part on the basis of data collected by one or more of the vehicle's sensors. The result of the assessment may be used as the basis for decisions about vehicle operation carried out by an autonomous driving module.

Growing community of inventors

North York, Canada

Elmira Amirloo Abolfathi