Articulated Object Tracking With Pseudo-Observable Segment Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional planning systems in vehicles, such as autonomous vehicles, face challenges in accurately representing and tracking articulated objects, like tractor-trailers, due to their varying footprints as they rotate, leading to overestimation of object extents and potentially cautious or inefficient navigation.
Innovation Solution
A multi-segment model is used to represent articulated objects, comprising geometric shapes for each segment, with iterative updates to estimate their track, utilizing extended Kalman filtering algorithms to correct errors between projected and measured states, and incorporating pseudo-observables to constrain segment motion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single bounding box is used to represent articulated objects, then the representation is simple, but the accuracy of object extents is poor leading to overestimation
Solution Approach 1:
The articulated object is divided into multiple rigid body segments (e.g., tractor and trailer segments) that can independently rotate and translate. Each segment is represented by its own bounding box with specific pose parameters (position, orientation, extent), allowing accurate representation of the varying footprint as segments rotate relative to each other, thereby resolving the overestimation problem of single bounding box representation.
2Measurement precision
If multi-segment model is used to represent articulated objects, then the accuracy of tracking is improved, but the computational complexity increases
Solution Approach 1:
The system employs an iterative tracking process where the pose of each segment is predicted based on previous state and motion models, then refined using sensor measurements and pseudo-observables. The extended Kalman filter provides feedback by comparing predicted and measured states, correcting errors in segment position and orientation estimates, thereby maintaining high tracking accuracy while managing computational complexity through efficient iterative refinement.
Solution Approach 2:
Pseudo-observables are introduced as intermediary constraints that relate the poses of different segments without requiring direct measurement of all parameters. These pseudo-observables encode the kinematic relationships between segments (e.g., fixed distance between connection points, relative orientation constraints), allowing the system to infer unmeasured segment states from available sensor data, thus reducing the computational burden of full-state estimation.
3Reliability
If conventional bounding box representation is used, then the system is computationally efficient, but navigation safety is compromised due to overestimation of object extents
Solution Approach 1:
The representation model transitions from static to dynamic by allowing each segment's bounding box parameters (position, orientation, extent) to change over time based on measured pose and predicted motion. The system continuously updates segment configurations as the articulated object moves and rotates, providing real-time accurate extent information that improves navigation safety while maintaining efficiency through optimized dynamic tracking algorithms.
Data Source
AI summary
Navigation systems can identify objects in an environment and generate representations of those objects. A representation of an articulated vehicle can include two segments rotated relative to each other about a pivot, with a first segment corresponding to a first portion of the articulated vehicle and the second segment corresponding to a second portion of the articulated vehicle. The articulated object can be tracked in the environment by generating estimated updated states of the articulated agent based on previous states and/or measured states of the object using differing motion model updates for the differing portions. The estimated updated states may be determined using one or more filtering algorithms, which may be constrained using pseudo-observables.