Egospace Data Structure for UAV Obstacle Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Micro air vehicles (MAVs) face challenges in high-speed autonomous flight in cluttered environments due to power and size restrictions, limiting sensing and computational resources for obstacle detection and motion planning, which existing technologies have not adequately addressed.
Innovation Solution
The development of an egospace data structure that compactly represents range measurements from multiple sensors, enabling efficient motion planning and obstacle avoidance by generalizing depth image obstacle representations and allowing for temporal fusion of depth maps, thereby improving collision-checking and trajectory generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If vision-based obstacle detection is used to satisfy sensing constraints, then power requirements are reduced and weight is minimized, but computational resources for real-time processing in cluttered environments become insufficient
Solution Approach 1:
The patent segments the environment representation into discrete depth bins along the flight path, creating a simplified one-dimensional depth map rather than processing full 3D point clouds. This segmentation reduces computational complexity while maintaining essential obstacle detection capabilities for real-time operation
Solution Approach 2:
The patent extracts only the critical depth information needed for collision avoidance by projecting 3D range data onto a one-dimensional depth map along the vehicle's forward axis. This extraction removes unnecessary spatial details while preserving the essential obstacle distance information required for real-time decision-making
2Reliability
If binocular stereo vision is used to populate occupancy grid world models, then forward-looking obstacle detection is achieved, but computational load and memory requirements increase
Solution Approach 1:
The patent transforms 3D spatial occupancy grid data into a one-dimensional depth map representation along the flight path. This dimensionality reduction projects complex three-dimensional obstacle information onto a single axis, dramatically reducing computational and memory requirements while maintaining the ability to detect forward obstacles
3Productivity
If disparity space representation is used for fast collision-checking, then trajectory evaluation speed is improved, but the field of regard is limited and sensor fusion becomes complex
Solution Approach 1:
The patent creates a universal depth map representation that serves multiple functions: it enables fast collision checking like disparity space, accommodates any sensor type (stereo, monocular, active range sensors), and provides a common framework for sensor fusion. The one-dimensional depth map along the flight path becomes a multi-functional data structure that simplifies rather than complicates sensor integration
4Reliability
If traditional motion planning methods are used in configuration space, then collision-free paths can be generated, but computational time increases and real-time performance is compromised
Solution Approach 1:
The patent extracts the essential collision information into a simplified one-dimensional depth map that directly represents obstacles along the flight path. This extraction eliminates the need for complex configuration space computations by providing direct depth measurements, enabling real-time collision-free path generation through simple comparison operations
Solution Approach 2:
The patent changes the representation parameters from full 3D configuration space coordinates to a one-dimensional depth parameter along the flight path. This parameter transformation simplifies the motion planning problem by reducing the dimensionality of the search space while maintaining the ability to generate collision-free trajectories
Data Source
AI summary
A method, device, framework, and system provide the ability to control an unmanned aerial vehicle (UAV) to avoid obstacle collision. Range data of a real-world scene is acquired using range sensors (that provide depth data to visible objects). The range data is combined into an egospace representation (consisting of pixels in egospace). An apparent size of each of the visible objects is expanded based on a dimension of the UAV. An assigned destination in the real world scene based on world space is received and transformed into egospace coordinates in egospace. A trackable path from the UAV to the assigned destination through egospace that avoids collision with the visible objects (based on the expanded apparent sizes of each of the visible objects) is generated. Inputs that control the UAV to follow the trackable path are identified.