Autonomous Vehicle Trajectory Planning for Occluded Actors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in navigating through environments with occlusions due to limited sensor capabilities, leading to incomplete information about their surroundings, which can trigger unnecessary operator takeovers and stops, and existing solutions fail to adequately model complex spatial and temporal dependencies of occluded regions.
Innovation Solution
A particle filter approach is used to model the unknown states of occluded regions by representing the environment as a mixture of hypotheses, each with a probability, which are updated using sensor data and motion models to generate trajectories for the vehicle, accounting for potential actors and their behaviors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If sensors are used to detect objects in the environment, then the autonomous vehicle can perceive its surroundings, but sensors cannot collect complete information due to occlusions from buildings, parked cars and other obstacles
Solution Approach 1:
The patent creates virtual copies (particles) of potential occluded actors to represent hypotheses about objects that cannot be directly observed. Each particle is a computational copy representing a possible state of an occluded object, allowing the system to reason about multiple potential scenarios simultaneously without requiring direct sensor observation of the actual objects.
Solution Approach 2:
The patent introduces particles as intermediary representations between the sensor data and the navigation decision-making process. These particles serve as mediators that carry probabilistic information about occluded regions, enabling the system to handle incomplete sensor information by processing these intermediate hypothesis representations rather than dealing directly with the uncertainty of occluded objects.
2Productivity
If the autonomous vehicle navigates through occluded regions, then it can reach destinations, but it may encounter unseen objects leading to safety risks
Solution Approach 1:
The patent performs preliminary actions by initializing particles in occluded regions before the vehicle actually encounters potential objects. The system proactively generates hypothesis particles representing possible occluded actors and their trajectories, allowing it to prepare navigation decisions in advance while accounting for potential unseen obstacles, rather than reacting only when objects are directly detected.
Solution Approach 2:
The patent implements feedback mechanisms where particle probabilities are continuously updated based on sensor observations and vehicle motion. As the vehicle moves and sensors gather new information, the system feeds this data back into the particle filter, updating the likelihood of each particle representing a real object, thereby refining navigation decisions through iterative feedback loops that balance exploration and safety.
3Adaptability or versatility
If traditional sensor-based detection is used, then the system can identify visible objects, but it fails to reason about occluded regions and cannot pass around blind corners
Solution Approach 1:
The patent applies dynamics by making the particle representation system adaptive and evolving over time. Particles are dynamically initialized, updated, and pruned based on changing sensor observations and vehicle motion. The system transitions from static sensor detection to dynamic hypothesis tracking, where particles evolve probabilistically to reflect new information, enabling the vehicle to adapt its understanding of occluded regions as it moves through the environment.
Solution Approach 2:
The patent adds a probabilistic hypothesis dimension to the traditional sensor detection space. Instead of only working with concrete sensor detections in physical space, the system introduces a fourth dimension of probabilistic belief about occluded objects. This dimensional extension allows the vehicle to reason about objects not present in the direct sensor field of view, transforming the problem from 3D spatial detection to 4D probabilistic reasoning that includes occluded regions.
Data Source
AI summary
An autonomous vehicle navigates an environment in which occlusions block the vehicle's ability to detect moving objects. The vehicle handles this by receiving sensor data corresponding to the environment, identifying one or more particles of a plurality of initialized particles that have a non-zero probability of being associated with a potential occluded actor in the occluded region, and generating a trajectory of the autonomous vehicle for traversing the environment based on spatiotemporal reasoning about the occluded region and taking into account the one or more particles. Each particle may be associated with a potential actor.