Robotic visual perception system

摘要

Described is a robotic visual perception system for determining a position and pose of a three-dimensional object. The system receives an external input to select an object of interest. The system also receives visual input from a sensor of a robotic controller that senses the object of interest. Rotation-invariant shape features and appearance are extracted from the sensed object of interest and a set of object templates. A match is identified between the sensed object of interest and an object template using shape features. The match between the sensed object of interest and the object template is confirmed using appearance features. The sensed object is then identified, and a three-dimensional pose of the sensed object of interest is determined. Based on the determined three-dimensional pose of the sensed object, the robotic controller is used to grasp and manipulate the sensed object of interest.

主权项

1. A visual perception system for determining a position and pose of a three-dimensional object, the system comprising:
one or more processors and a memory having instructions such that when the instructions are executed, the one or more processors perform operations of: receiving visual input from a sensor of a robotic controller that senses an object of interest; extracting a plurality of rotation-invariant shape features from the sensed object of interest; extracting a plurality of rotation-invariant shape features from a set of object templates; identifying a match between the sensed object of interest and an object template using shape features; confirming the match between the sensed object of interest and the object template using appearance features; identifying the sensed object of interest and determining a three-dimensional pose of the sensed object of interest; and using the robotic controller to handle the sensed object of interest based on the determined three-dimensional pose of sensed object of interest.