Autonomous Vehicle Control via Sensorimotor Primitive Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle control systems rely on high-definition maps and high-precision GPS, which are unreliable in areas without network connectivity and require complex neural networks that are difficult to train and validate, especially for handling unknown driving environments.
Innovation Solution
The system employs a high-level controller that selects and prioritizes sensorimotor primitive modules to generate vehicle trajectory and speed profiles using a scene understanding module, arbitration module, and vehicle control module, which processes sensor data to control the autonomous vehicle without the need for HD maps or high-precision GPS, utilizing neuromorphic control models and dynamic mapping modules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-definition maps and high-precision GPS are used to provide lane-level topology and vehicle location, then the autonomous vehicle can accurately navigate in known environments, but the system becomes unreliable in areas without network connectivity and cannot handle unknown driving environments
Solution Approach 1:
The patent segments the autonomous driving system into multiple independent modules (sensorimotor primitives) that can operate autonomously without requiring HD maps or high-precision GPS. Each primitive handles specific driving tasks (e.g., lane following, intersection navigation, parking) using basic sensor data, allowing the vehicle to navigate unknown environments by combining these modular functions rather than relying on a centralized mapping system.
2Adaptability or versatility
If a single end-to-end neural network is used to map image pixels to control actions, then the system can handle all driving scenarios, but the computational complexity and power consumption increase significantly
Solution Approach 1:
The patent divides the complex end-to-end neural network into multiple smaller, specialized sensorimotor primitive modules. Each primitive is a simplified neural network trained for a specific driving task (e.g., adaptive cruise control, lane keeping, turn execution). This segmentation reduces the computational burden and power consumption of each individual module while maintaining overall system versatility through modular composition.
Solution Approach 2:
The system dynamically selects and activates only the sensorimotor primitives relevant to the current driving scenario, rather than continuously running all possible functions. This dynamic module selection based on scene understanding reduces real-time computational complexity and power consumption while maintaining adaptability to handle diverse driving situations.
3Adaptability or versatility
If a single end-to-end neural network is used for all driving scenarios, then the system can learn new features, but system-level re-validation is required for any new features learned
Solution Approach 1:
The patent segments the validation process into module-level validation for each sensorimotor primitive rather than requiring complete system-level re-validation. Each primitive can be independently trained, tested, and validated for its specific function, allowing new features to be added to individual modules without re-validating the entire autonomous driving system, thus reducing validation complexity while maintaining adaptability.
4Measurement precision
If high-precision GPS is used to locate the vehicle, then the vehicle position can be accurately determined, but the GPS is not available in certain areas such as those with less satellite visibility
Solution Approach 1:
The patent enables the autonomous vehicle to determine its position and navigate using only onboard sensors (cameras, lidars, radars) and basic maps, making the system self-sufficient without external GPS infrastructure. The sensorimotor primitives process sensor data to infer vehicle state and environment, allowing the vehicle to operate independently in GPS-denied environments like urban canyons or areas with satellite visibility issues.
Data Source
AI summary
Systems and methods are provided for controlling an autonomous vehicle (AV). A scene understanding module of a high-level controller selects a particular combination of sensorimotor primitive modules to be enabled and executed for a particular driving scenario from a plurality of sensorimotor primitive modules. Each one of the particular combination of the sensorimotor primitive modules addresses a sub-task in a sequence of sub-tasks that address a particular driving scenario. A primitive processor module executes the particular combination of the sensorimotor primitive modules such that each generates a vehicle trajectory and speed profile. An arbitration module selects one of the vehicle trajectory and speed profiles having the highest priority ranking for execution, and a vehicle control module processes the selected one of vehicle trajectory and speed profiles to generate control signals used to execute one or more control actions to automatically control the AV.