Autonomous Driving Decision Feedback System for Planning Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving vehicles often make imperfect decisions due to limitations in their planning and control systems, leading to uncomfortable rides, traffic rule violations, risky situations, and longer trip times, as they struggle to execute decisions effectively and efficiently.
Innovation Solution
A system that includes a perception and planning module to generate driving decisions, a logging module to record sensor outputs and control inputs, and a decision problem recognition module using supervised learning to categorize and improve driving decisions, with local and global improvement modules using reinforcement learning and machine learning to optimize safety, comfort, and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If the ADV decision system makes autonomous driving decisions, then the vehicle can navigate automatically with minimal human interaction, but the decisions may not be perfect leading to uncomfortable rides, traffic rule violations, risky situations, or longer trip times
Solution Approach 1:
The patent implements a feedback mechanism where planning module feedback is used to identify decision problems in the ADV decision system. The system continuously monitors decision outcomes and uses this feedback to improve future decisions, addressing the reliability issue while maintaining automation.
Solution Approach 2:
The system employs self-service through automated decision problem identification and improvement modules that autonomously analyze and correct decision errors without human intervention, enabling the system to self-optimize while maintaining high automation levels.
2Speed
If the ADV planning and control system executes decisions, then the vehicle can respond to surrounding objects, but it may not satisfy one or more decision parameters causing incomplete execution
Solution Approach 1:
The planning module provides feedback on decision parameter satisfaction, allowing the system to identify and correct incomplete executions. This feedback loop ensures that both response speed and parameter precision are maintained simultaneously.
Solution Approach 2:
The system performs preliminary validation of decision parameters before execution, ensuring that all required parameters are satisfied in advance, which prevents incomplete execution while maintaining rapid response to objects.
3Reliability
If the ADV decision system prioritizes safety and rule compliance, then traffic violations and risky situations are reduced, but trip time may increase
Solution Approach 1:
The system dynamically adjusts decision parameters based on the situation, allowing it to maintain safety and compliance while optimizing for reasonable trip times. By changing parameters adaptively rather than using fixed conservative values, the system balances safety with efficiency.
Solution Approach 2:
The decision-making system transitions from static rule-following to dynamic optimization, where safety constraints are maintained but other parameters can be adjusted in real-time to minimize trip time, creating a balanced approach to autonomous driving.
4Object-affected harmful factors
If the ADV decision system makes conservative decisions, then safety and comfort are improved, but trip time and efficiency are reduced
Solution Approach 1:
The system dynamically adjusts the conservatism level of decisions based on real-time conditions and learned patterns. Rather than always being conservative, the system adapts its behavior to maintain safety and comfort while improving efficiency through experience.
Solution Approach 2:
The system changes decision parameters based on accumulated knowledge and feedback, allowing it to maintain protective levels for safety and comfort while optimizing efficiency parameters to reduce unnecessary conservatism and improve productivity.
Data Source
AI summary
In one embodiment, systems and methods are disclosed for a planning-driven framework for an driving vehicle (ADV) driving decision system. Driving decisions are classified into at least seven categories, including: conservative decision, aggressive decision, conservative parameters, aggressive parameters, early decision, late decision, and non-decision problem. Using the outputs of an ADV decision planning module, an ADV driving decision problem is identified, categorized, and diagnosed. A local driving decision improvement can be determined and executed in a short time frame on the ADV. For a long term solution, if needed, the driving decision problem can be uploaded to an analytics server. The driving decision problems from a large plurality of ADVs can be aggregated and analyzed for improving the ADV decisions system for all ADVs.