Image-Space Trajectory Planning for Uncertain Depth Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in navigating environments with unreliable depth estimates due to complex-shaped objects like trees with intermittent foliage, leading to uncertain and risky navigation paths.
Innovation Solution
Image space motion planning techniques identify regions with low confidence depth estimates and associate them with higher collision risk, optimizing the vehicle's trajectory to avoid these areas by projecting the planned path into the image space and minimizing collision risk.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the autonomous vehicle uses visual odometry to estimate position and orientation based on captured images, then the navigation system can guide the vehicle through the physical environment, but the depth estimates become unreliable in regions with complex-shaped objects like trees with intermittent foliage
Solution Approach 1:
The patent segments the physical environment into multiple regions based on depth estimation reliability. Regions with complex-shaped objects (e.g., trees with intermittent foliage) are identified and separated from regions with reliable depth estimates. This segmentation allows the navigation system to treat different regions differently, avoiding unreliable regions while utilizing reliable regions for path planning.
Solution Approach 2:
The patent introduces an intermediary classification layer between image capture and trajectory planning. This intermediary system classifies regions as either reliable or unreliable for depth estimation, acting as a mediator that translates raw image data into actionable navigation information. The classification map serves as an intermediary representation that guides the trajectory optimization process.
2Reliability
If the autonomous vehicle plans a trajectory to avoid regions with low confidence depth estimates, then the collision risk is reduced, but the navigation path may become more conservative and less efficient
Solution Approach 1:
The patent dynamically adjusts the trajectory planning based on real-time classification of reliable and unreliable regions. Rather than using a fixed conservative path, the system continuously adapts the navigation trajectory to exploit available reliable regions while avoiding unreliable ones. This dynamic approach allows the vehicle to maintain safety while optimizing for efficiency by taking advantage of favorable environmental conditions.
Solution Approach 2:
The patent changes the parameter space of trajectory planning by incorporating region reliability classifications as additional constraints and optimization criteria. The cost function for trajectory optimization is modified to account for depth estimation reliability, transforming the planning problem from simple path finding to reliability-aware path optimization. This parameter change enables the system to balance safety and efficiency by quantifying and optimizing over region reliability.
Data Source
AI summary
An autonomous vehicle that is equipped with image capture devices can use information gathered from the image capture devices to plan a future three-dimensional (3D) trajectory through a physical environment. To this end, a technique is described for image-space based motion planning. In an embodiment, a planned 3D trajectory is projected into an image-space of an image captured by the autonomous vehicle. The planned 3D trajectory is then optimized according to a cost function derived from information (e.g., depth estimates) in the captured image. The cost function associates higher cost values with identified regions of the captured image that are associated with areas of the physical environment into which travel is risky or otherwise undesirable. The autonomous vehicle is thereby encouraged to avoid these areas while satisfying other motion planning objectives.