Autonomous Low-Speed Maneuver Replay Without Digital Road Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automotive autonomous driving systems face challenges in optimizing performance for recurrent low-speed maneuvers in digital road maps-free areas, where obstacles and surrounding vehicles create unpredictable paths, leading to inefficiencies and potential errors due to decreased driver attention during routine maneuvers.
Innovation Solution
An automotive autonomous driving system with a collaborative machine learning approach that uses a database of recurrent low-speed maneuvers, including geolocation data and sensory system inputs, to compute and update autonomous driving paths in real-time, allowing for adaptive navigation and optimization through a remote service center, enabling the system to handle new obstacles and constraints without re-storing maneuvers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If autonomous driving systems use traditional path planning methods in digital road maps-free areas, then the system can operate without map data, but the performance and accuracy of recurrent low-speed maneuvers deteriorate due to unpredictable obstacles and vehicles
Solution Approach 1:
The system pre-stores recurrent low-speed maneuvers and their associated path data in a database before actual execution. This preliminary storage of maneuver patterns, spatial constraints, and obstacle information enables the system to quickly retrieve and execute optimized paths during actual operations without requiring real-time map data, thus improving both reliability and performance in digital road maps-free areas.
Solution Approach 2:
The system implements a feedback mechanism where maneuver data from multiple vehicles is collected and used to update and refine the global navigation optimization policy. This continuous feedback loop allows the system to learn from collective experiences, improve path planning accuracy over time, and adapt to new obstacle patterns, thereby enhancing the reliability of recurrent low-speed maneuvers without requiring digital road maps.
2Productivity
If the system stores and re-executes predefined maneuver paths, then efficiency improves for recurrent maneuvers, but the system cannot adapt to new obstacles or environmental changes
Solution Approach 1:
The system dynamically adapts stored maneuver paths by integrating real-time sensory data with pre-stored path information. When new obstacles or environmental changes are detected during execution, the system modifies the predefined paths on-the-fly by combining them with current sensor inputs, thus maintaining both the efficiency of recurrent maneuvers and the adaptability to new conditions.
Solution Approach 2:
The maneuver database serves multiple functions: it stores predefined paths for efficient execution, provides training data for machine learning models, and acts as a reference for real-time path adjustments. This multi-functionality allows the system to maintain high efficiency for recurrent maneuvers while simultaneously adapting to new obstacles through sensory integration and collaborative learning from multiple vehicles.
3Reliability
If the system uses collaborative machine learning with multiple vehicles, then navigation optimization improves through collective learning, but system complexity and data processing requirements increase
Solution Approach 1:
The system introduces a remote service center as an intermediary that centralizes the collection, processing, and analysis of maneuver data from multiple vehicles. This intermediary handles the complex tasks of training machine learning models and generating global navigation optimization policies, while individual vehicles only need to execute relatively simple local policies and send data uploads, thus improving navigation accuracy without proportionally increasing on-vehicle system complexity.
Solution Approach 2:
The collaborative learning system is segmented into distinct functional components: local maneuver execution units in each vehicle, data collection modules, remote training servers, and policy deployment mechanisms. This segmentation allows each component to be optimized independently and simplifies the overall system architecture by distributing computational loads, thereby enabling complex collaborative learning while managing system complexity through modular design.
4Adaptability or versatility
If the system real-time computes autonomous driving paths using sensory data, then adaptability to new obstacles improves, but computational time and processing power requirements increase
Solution Approach 1:
The system pre-computes and stores optimized maneuver paths and spatial constraint data in a maneuver database before actual execution. During real-time operation, the system retrieves these pre-computed paths and performs only minor adjustments based on current sensory data, rather than computing complete paths from scratch. This preliminary action significantly reduces real-time computational time while maintaining adaptability to new obstacles through selective path modification.
Solution Approach 2:
The system applies different processing strategies to different parts of the navigation problem: it uses pre-stored paths for stable, predictable segments of the maneuver, and applies real-time sensory-based adjustments only to portions of the path affected by new obstacles or constraints. This local quality approach minimizes overall computational time by avoiding redundant calculations in unchanged areas while maintaining adaptability where needed.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An automotive autonomous driving system (1) configured to cause a motor vehicle (2) to perform recurrent low speed manoeuvres in autonomous driving in digital road maps-free. The automotive autonomous driving system (1) is configured to store recurrent low speed manoeuvres in the form of manoeuvre data descriptive of the environments in which the recurrent low speed manoeuvres are to be repeated in autonomous driving and comprising geolocation data representative of starting and end points of the recurrent low speed manoeuvres and of spatial constraints, obstacles, and associated dimensions, identified in the environments where the recurrent low speed manoeuvres are to be repeated in autonomous driving, when they were stored. The automotive autonomous driving system (1) is designed to cause the motor-vehicle (2) to repeat stored recurrent low speed manoeuvres by causing the motor-vehicle (2) to drive along an autonomous driving path computed, along with associated longitudinal and lateral dynamics, based on both stored data descriptive of the environments in which the recurrent low speed manoeuvres are to be performed in autonomous driving and on data received from an automotive sensory system, whereby the autonomous driving path and the associated longitudinal and lateral dynamics are computed by taking account of both the spatial constraints, obstacles and their dimensions identified in the environment where the recurrent low speed manoeuvre is to be repeated in autonomous driving identified when it was stored, and of any new spatial constraints and/or obstacles appeared in the environment in which the recurrent low speed manoeuvre is to be performed after it was stored and which can be identified based on the data received from the automotive sensory system. The automotive autonomous driving system (1) is further designed to optimize repetition in autonomous driving of stored recurrent low speed manoeuvres based on a collaborative navigation optimization approach implemented by the automotive autonomous driving system (1) and a remote service centre (10).