Amazon Technologies, Inc.

Boshen Niu

Shreekant Gayaka

Eric Foxlin

Simon Edwards-Parton

Lu Xia

James Ballantyne

Harish Annavajjala

Shengchen Liu

An autonomous mobile device (AMD) moves within in a physical space. The AMD may encounter features in the physical space, such as flooring transitions, or obstacles that may impede movement. Sensors such as an inertial measurement unit (IMU), wheel encoders, and motor torque sensors acquire sensor data. Based on the sensor data, measured motion values are determined. Based on physical parameters of the AMD and the sensor data, predicted motion values are determined. A residual or difference between the predicted and measured motion values is calculated. If the residual exceeds a threshold range, a collision may be deemed to have occurred. Characteristics of the residual may be used to determine which side of the AMD collided with an obstacle. Collisions are reliably detected while false detections due to features such as uneven floors or flooring transitions are reduced or eliminated.

System to determine collisions of an autonomous mobile device

Described herein is a system for improving dynamic object detection by including posture data. For example, the system may build an occupancy map representing objects present in an environment while tracking dynamic objects that should not be included in the occupancy map (e.g., people, pets, or other mobile objects). The system may classify a user as a dynamic object based on a posture indicated by Visual Perception (VP) data. For example, the VP data may distinguish between first postures (e.g., standing, walking, etc.) and second postures (e.g., sitting, laying down, etc.). Thus, the system can classify users associated with the first postures as dynamic objects to be removed from the occupancy map, while including users associated with the second postures in the occupancy map. Thus, the device removes humans from the occupancy map without removing furniture.

Dynamic object detection using posture data

A physical space contains stationary objects that do not move over time (e.g., a couch) and may have non-stationary objects that do move over time (e.g., people and pets). An autonomous mobile device (AMD) determines and uses an occupancy map of stationary objects to find a route from one point to another in a physical space. Non-stationary objects are detected and prevented from being incorrectly added to the occupancy map. Point cloud data is processed to determine first candidate objects. Image data is processed to determine second candidate objects. These candidate objects are associated with each other and their characteristics assessed to determine if the candidate objects are stationary or non-stationary. The occupancy map is updated with stationary obstacles. During navigation, the occupancy map may be used for route planning while the non-stationary objects are used for local avoidance.

System to determine non-stationary objects in a physical space

An autonomous mobile device (AMD) may move around a physical space while performing tasks. Sensor data is used to determine an occupancy map of the physical space. Some objects within the physical space may be difficult to detect because of characteristics that result in lower confidence in sensor data, such as transparent or reflective objects. To include difficult-to-detect objects in the occupancy map, image data is processed to identify portions of the image that includes features associated with difficult-to-detect objects. Given the portion that possibly includes difficult-to-detect objects, the AMD attempts to determine where in the physical space that portion corresponds to. For example, the AMD may use stereovision to determine the physical area associated with the features depicted in the portion. Objects in that area are included in an occupancy map annotated as objects that should persist unless confirmed to not be within the physical space.

Growing community of inventors

San Mateo, CA, United States of America

Boshen Niu