Neurally-Inspired Robot Perception System for Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current robotic systems face challenges in segregating discrete objects of interest from their background, require separate subsystems for object segregation, recognition, and tracking, and struggle with maintaining temporal continuity of objects in dynamic environments, while also separating system learning from system use.
Innovation Solution
A unified mechanism using a neurally-inspired mathematical model with spatial attention and semantics modules to fuse disparate sensory information, enabling online learning and tracking of objects in real-time, mimicking human perception by allocating sensor resources efficiently and dynamically.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate subsystems are used for object segregation, recognition, and tracking, then each function can be specialized, but the device complexity increases and modules operate in isolation with little interaction
Solution Approach 1:
The patent merges object segregation, recognition, and tracking into a unified neural processing system where a single neural network performs all three functions simultaneously through integrated processing pathways, eliminating the need for separate interacting subsystems
Solution Approach 2:
The neural network is designed as a universal processing system that handles multiple object analysis functions (segregation, recognition, tracking) within a single architecture, allowing one system to perform what previously required multiple specialized subsystems
2Measurement precision
If learning is treated as an off-line method in a separate time frame, then learning can be performed systematically, but it interrupts system operation and reduces productivity
Solution Approach 1:
The neural network enables continuous learning by processing sensory inputs and updating object representations in real-time as the robot operates, allowing learning to occur continuously during system operation rather than requiring separate offline training periods
Solution Approach 2:
The system pre-processes sensory information through the neural network during normal operation, continuously building and refining object representations in advance of when they are needed for recognition or tracking tasks
3Ease of manufacture
If stovepiped sensory processing is used with isolated modules, then each processing stage is simplified, but continuous fusion and learning of pertinent information cannot be achieved
Solution Approach 1:
The patent combines multiple sensory processing pathways (visual, auditory, and other sensor information) into a unified neural network that simultaneously processes and fuses information from all modalities, preventing information loss that occurs in isolated processing stages
Data Source
AI summary
Sensory processing of visual, auditory, and other sensor information (e.g., visual imagery, LIDAR, RADAR) is conventionally based on “stovepiped,” or isolated processing, with little interactions between modules. Biological systems, on the other hand, fuse multi-sensory information to identify nearby objects of interest more quickly, more efficiently, and with higher signal-to-noise ratios. Similarly, examples of the OpenSense technology disclosed herein use neurally inspired processing to identify and locate objects in a robot's environment. This enables the robot to navigate its environment more quickly and with lower computational and power requirements.