Layered Sensor Fusion for Reliable Autonomous Obstacle Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional sensor fusion techniques for autonomous mobile devices (AMDs) are complex and prone to failure in dynamic physical spaces, especially when new obstacles are introduced or sensors become unavailable.
Innovation Solution
A system that stores information from different sensor modalities in aligned layers representative of areas in the physical space, allowing for flexible addition and removal of modalities and aggregation profiles, which enables reliable obstacle detection and navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional sensor fusion techniques are used, then obstacle detection capability is provided, but system complexity increases and reliability decreases in dynamic environments
Solution Approach 1:
The patent segments sensor data processing into distinct layers (e.g., depth layer, semantic layer, fusion layer) where each layer handles specific aspects of sensor fusion. This modular segmentation reduces overall system complexity while maintaining reliable obstacle detection by allowing independent processing and validation at each layer.
Solution Approach 2:
The patent introduces intermediate data structures and processing stages that mediate between raw sensor data and final obstacle detection results. These intermediaries buffer and organize sensor inputs, reducing the direct complexity of fusion operations while improving reliability through systematic data transformation and validation steps.
2Adaptability or versatility
If traditional sensor fusion techniques are used, then sensor data is processed, but the system fails when sensors become unavailable or new obstacles are introduced
Solution Approach 1:
The patent implements dynamic sensor fusion where the system adapts its processing pipeline based on available sensors and environmental conditions. When sensors become unavailable, the system dynamically reconfigures which sensor layers are active, maintaining reliable obstacle detection with remaining sensors rather than failing completely.
Solution Approach 2:
The patent changes processing parameters dynamically based on sensor availability and obstacle characteristics. When new obstacles are introduced or sensors fail, the system adjusts fusion weights, thresholds, and active processing layers to maintain reliable detection across varying conditions.
3Adaptability or versatility
If flexible addition and removal of sensor modalities is implemented, then adaptability improves, but system complexity increases
Solution Approach 1:
The patent segments the sensor fusion system into independent, interchangeable modalities (e.g., depth modality, semantic modality, fusion modality) that can be added or removed without affecting the core architecture. Each segment is self-contained with standardized interfaces, allowing flexible configuration while managing complexity through modular design.
Solution Approach 2:
The patent creates a universal sensor fusion architecture where a single framework handles multiple sensor modalities through common data structures and processing pipelines. This multi-functional design allows different sensor types to be added or removed while reusing the same core fusion logic, reducing overall system complexity despite increased flexibility.
Data Source
AI summary
A physical space includes obstacles with different characteristics. Different sensors may detect some obstacles well, while other obstacles are poorly detected. An autonomous mobile device (AMD) uses data from various sensors to determine information about the physical space that is expressed in layers. Each layer represents specified areas within at least a portion of the physical space, with each area having a value representative of whether an obstacle is present. Some layers may represent specific volumes, or height ranges. For example, different layers may represent a low height, a medium height, and a high height. Aggregated data may be determined using the values from multiple layers. The aggregated data may be calculated using an aggregation profile to specify a relative weighting for the values in a particular layer. The aggregated data may then be processed to determine maps, such as a navigation map for autonomous movement, floorplan, and so forth.