Vehicle Navigation Using Occupancy Grids and Deep RL
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face inefficiencies and computational bottlenecks in processing sensor data for vehicles, leading to delays in making predictive decisions for navigating through environments while accounting for other objects.
Innovation Solution
A vehicle control system utilizing a deep reinforcement learning neural network to process occupancy grid maps, integrating sensor data from both the vehicle and infrastructure, to determine optimal actions for navigation, thereby reducing computational resource consumption and decision-making time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional sensor data processing methods are used, then comprehensive environmental perception is achieved, but computational resource consumption increases and decision-making time increases
Solution Approach 1:
The patent segments the sensor data processing by dividing the environment into discrete occupancy grid cells. Each cell is independently evaluated for occupancy status, allowing parallel processing and reducing the computational burden of analyzing continuous sensor data. This segmentation enables efficient resource usage while maintaining comprehensive environmental perception.
Solution Approach 2:
The system performs preliminary actions by pre-processing sensor data into occupancy grid maps before navigation decisions are required. The deep reinforcement learning neural network is trained in advance on simulated data, enabling it to make rapid decisions during actual operation without extensive real-time computation. This preliminary preparation reduces decision-making time while maintaining accuracy.
2Productivity
If deep reinforcement learning neural network is implemented, then decision-making time is reduced, but device complexity increases
Solution Approach 1:
The patent uses a simulated environment to create copies of real-world scenarios for training the deep reinforcement learning neural network. The simulation generates synthetic occupancy grid maps and corresponding correct actions, allowing the network to learn from numerous examples without requiring complex real-time data collection and annotation systems. This copying approach simplifies the actual deployment system.
Solution Approach 2:
The system replaces complex mechanical decision-making processes with a neural network-based computational approach. Instead of using traditional algorithms and rule-based systems that require extensive processing logic, the patent substitutes these with a trained neural network that makes decisions through learned patterns, reducing the complexity of the control system while improving decision speed.
3Measurement precision
If occupancy grid map processing is performed, then navigation accuracy is improved, but computational resource consumption increases
Solution Approach 1:
The system applies partial action by processing only the relevant portion of the occupancy grid map that contains obstacles or objects of interest, rather than analyzing the entire map. The neural network focuses computation on areas where navigation decisions are needed, reducing overall computational energy consumption while maintaining navigation accuracy in critical regions.
Data Source
AI summary
A portion of an occupancy grid for an area is obtained. The occupancy grid map is generated based on collected data of a host object in the area and collected data of respective target objects in the area. A predicted portion occupancy grid for the area is generated based on predicted data of the host object and predicted data of the respective target objects. An action is determined based on inputting the portion and the predicted portion of the occupancy grid to a deep reinforcement learning neural network. A host object is operated based on the action.