Embodied Intelligence Inc.

Xi Chen

Yan Duan

Yide Shentu

Nikhil Mishra

Haoran Tang

Andrew Amir Vaziri

Shiyao Wu

Maximilian Sieb

Ian Rust

Mostafa Rohaninejad

David Mascharka

Tianhao Zhang

Jasmine Deng

Carlos Florensa

Yu Xuan Liu

YuXuan Liu

Various embodiments of the present technology generally relate to robotic devices and artificial intelligence. More specifically, some embodiments relate to an artificial neural network training method that does not require extensive training data or time expenditure. The few-shot training model disclosed herein includes attempting to pick up items and, in response to a failed pick up attempt, transferring and generalizing information to similar regions to improve probability of success in future attempts. In some implementations, the training method is used to robotic device for picking items from a bin and perturbing items in a bin. When no picking strategies with high probability of success exist, the robotic device may perturb the contents of the bin to create new available pick-up points. In some implementations, the device may include one or more computer-vision systems.

Training artificial networks for robotic picking

Various embodiments of the present technology generally relate to robotic devices, artificial intelligence, and computer vision. More specifically, some embodiments relate to an imaging process for detecting failure modes in a robotic motion environment. In one embodiment, a method of detecting failure modes in a robotic motion environment comprises collecting one or more images of a multiple scenes throughout a robotic motion cycle. Images may be collected by one or more cameras positioned at one or more locations for collecting images with various views. Images collected throughout the robotic motion cycle may be processed in real-time to determine if any failure modes are present in their respective scenes, report when failure modes are present, and may be used to direct a robotic device accordingly.

Imaging process for detecting failures modes

Various embodiments of the technology described herein generally relate to systems and methods for trajectory optimization with machine learning techniques. More specifically, certain embodiments relate to using neural networks to quickly predict optimized robotic arm trajectories in a variety of scenarios. Systems and methods described herein use deep neural networks to quickly predict optimized robotic arm trajectories according to certain constraints. Optimization, in accordance with some embodiments of the present technology, may include optimizing trajectory geometry and dynamics while satisfying a number of constraints, including staying collision-free and minimizing the time it takes to complete the task.

Trajectory optimization using neural networks

Various embodiments of the present technology generally relate to robotic devices and artificial intelligence. More specifically, some embodiments relate to a robotic device for picking items from a bin and perturbing items in a bin. In some implementations, the device may include one or more computer-vision systems. A computer-vision system, in accordance with the present technology, may use at least two two-dimensional images to generate three-dimensional (3D) information about the bin and items in the bin. Based on the 3D information, a strategy for picking up items from the bin is determined. When no strategies with high probability of success exist, the robotic device may perturb the contents of the bin to create new available pick-up points and re-attempt to pick up an item.

Systems and methods for robotic picking

Various embodiments of the present technology generally relate to robotic devices and artificial intelligence. More specifically, some embodiments relate to modeling uncertainty in neural network segmentation predictions of imaged scenes having a plurality of objects. In some embodiments, a computer vision system for guiding robotic picking utilizes a method for uncertainty modeling that comprises receiving one or more images of a scene comprising a plurality of distinct objects, generating a plurality of segmentation predictions each comprising one or more object masks, identifying a predefined confidence requirement, wherein the confidence requirement identifies a minimum amount of required agreement for a region, and outputting one or more object masks based on the confidence requirement. The systems and methods disclosed herein leverage the use of a plurality of hypotheses to create a distribution of possible segmentation outcomes in order model uncertainty associated with image segmentation.

Confidence-based segmentation of multiple units

Various embodiments of the technology described herein generally relate to systems and methods for trajectory optimization with machine learning techniques. More specifically, certain embodiments relate to using neural networks to quickly predict optimized robotic arm trajectories in a variety of scenarios. Systems and methods described herein use deep neural networks to quickly predict optimized robotic arm trajectories according to certain constraints. Optimization, in accordance with some embodiments of the present technology, may include optimizing trajectory geometry and dynamics while satisfying a number of constraints, including staying collision-free, and minimizing the time it takes to complete the task.

Various embodiments of the present technology generally relate to robotic devices, computer-vision systems, and artificial intelligence. More specifically, some embodiments relate to a computer-vision system for robotic devices. In some implementations, a computer-vision is coupled to a robotic arm for picking items from a bin and perturbing items in a bin. A computer-vision system, in accordance with the present technology, may use at least two two-dimensional (2D) images to generate three-dimensional (3D) information about the bin and items in the bin. A method of training artificial neural networks to generate 3D information from 2D images includes obtaining ground truth data for a scene, obtaining at least two images of the scene, and providing the ground truth data and the at least two images to the artificial neural network configured to generate a depth map from the images based on the ground truth data.

Determine depth with pixel-to-pixel image correspondence for three-dimensional computer vision

Various embodiments of the present technology generally relate to robotic devices and artificial intelligence. More specifically, some embodiments relate to a robotic device for picking items from a bin and perturbing items in a bin. The robotic device may include one or more picking elements and one or more perturbation elements for disturbing a present arrangement of items in the bin. In an exemplary embodiment, a perturbation element comprises a compressed air valve. In some implementations, the robotic device may also include one or more computer-vision systems. Based on image data from the one or more computer-vision systems, a strategy for picking up items from the bin is determined. When no strategies with high probability of success exist, the robotic device may perturb the contents of the bin to create new available pick-up points.

Systems and methods for robotic picking and perturbation

Various embodiments of the present technology generally relate to robotic devices and artificial intelligence. More specifically, some embodiments relate to an artificial neural network training method that does not require extensive training data or time expenditure. The few-shot training model disclosed herein includes attempting to pick up items and, in response to a failed pick up attempt, transferring and generalizing information to similar regions to improve probability of success in future attempts. In some implementations, the training method is used to robotic device for picking items from a bin and perturbing items in a bin. When no picking strategies with high probability of success exist, the robotic device may perturb the contents of the bin to create new available pick-up points. In some implementations, the device may include one or more Computer-vision systems.

Growing community of inventors

Berkeley, CA, United States of America

Xi Chen