Anthropomorphic Robot Path Learning via 3D Vision Imitation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anthropomorphic robot programming technologies require long times and specialized programmers, and existing teaching methods risk damaging actual tools during training.
Innovation Solution
A computer-implemented method and system that allows anthropomorphic robots to learn complex paths by imitating human operator movements, using three-dimensional vision sensors to capture and process images, calculate roto-translation matrices, and simulate trajectories, enabling automatic learning and reduced programming time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional field point learning or offline path programming is used for robot programming, then the robot can be programmed to perform work, but the programming time is long and specialized programmers are required
Solution Approach 1:
The system captures the operator's actual movements using 3D vision sensors and creates a digital copy of the trajectory. This copied trajectory is then processed and transferred to the robot, eliminating the need for traditional programming methods. The copying principle directly resolves the contradiction by replacing time-consuming programming with rapid movement capture and replication.
Solution Approach 2:
The patent replaces the mechanical programming process (manual teaching or offline path programming) with an optical system (3D vision sensors). The vision system captures movements optically and processes them computationally, substituting the traditional mechanical programming approach with a faster optical-digital system that dramatically reduces programming time.
2Reliability
If traditional teaching methods are used to train the robot, then the robot learns the path, but there is a risk of damaging actual tools during training
Solution Approach 1:
The system performs preliminary capture of the operator's movements before any robot execution occurs. The 3D vision sensors record the complete trajectory during the operator's demonstration, and this recorded data is processed and validated before the robot begins execution. This preliminary action ensures that the robot learns from a pre-captured, safe demonstration without risking tool damage during the learning process.
Solution Approach 2:
The patent introduces an intermediary system (3D vision sensors and processing unit) between the operator's movements and the robot's execution. This intermediary captures, processes, and validates the trajectory data, acting as a buffer that eliminates direct risk to tools during training. The intermediary transforms the teaching process from direct physical interaction to indirect digital replication, ensuring tool safety.
3Adaptability or versatility
If complex paths are programmed for the robot, then the robot can perform finishing operations, but the programming process is time-consuming and requires expertise
Solution Approach 1:
The system enables the robot to self-learn complex paths through the operator's natural movements. Instead of requiring the operator to program complex trajectories manually or provide detailed instructions, the operator simply demonstrates the desired path, and the system automatically captures and processes the movement data. This self-service approach allows the robot to acquire complex path information independently, reducing programming complexity while maintaining full capability for finishing operations.
Data Source
Figure 1
Figure 2~3
AI summary
The computer-implemented method (M) for self-learning of an anthropomorphic robot comprises the following steps: acquisition at predefined time intervals and by means of a three-dimensional vision sensor (V) of a sequence of images (I) relating to a machining operation carried out by an operator (O) on a reference object (R) within a learning area (step 12); processing, for each of the acquired images (I) of said sequence of images, of a point cloud (N) relating to the operator (O) and to the reference object (R) within the framed learning area (step 13); for each processed point cloud (N), identification of a tool (U) used by the operator (O) during the machining operation and of different parts of the body of the operator (O) (step 21); determination of coordinates (X, Y, Z) of movement axes relating to the tool (U) and to the different parts of the body of the operator (O) (step 22); calculation of a roto-translation matrix (RX, RY, RZ) of the tool (U) in space starting from the coordinates (X, Y, Z) of movement axes relating to the tool (U) and to different parts of the body of the operator (O) (step 31); calculation of the complete trajectory (T) of the tool (U) starting from the succession of the coordinates (X, Y, Z) of the movement axes of the tool (U) and from the succession of the angles of the roto-translation matrix (RY, RY, RZ) that identify the spatial inclination of the movement axes of the tool (U); conversion of the calculated trajectory (T) into a program module (P) for an anthropomorphic robot provided with a tool (U), wherein the program module (P) comprises a sequence of instructions for the movement of the tool (U) along the trajectory (T).