Vehicle Predictive Control Using Safe Terminal State Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive optimization methods for vehicle control, such as model predictive control (MPC), are computationally intense and limited to short prediction horizons, leading to feasibility issues and increased risk of unfeasible vehicle states.
Innovation Solution
Constrain predictive optimization with a set of allowed vehicle states at a short first prediction horizon, determined to result in safe vehicle states at a longer second horizon, using offline processing and machine learning to enhance computational efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If predictive optimization is used with a longer prediction horizon to ensure safe vehicle states, then the reliability of vehicle control is improved, but the computational complexity increases making it infeasible for real-time control
Solution Approach 1:
The prediction horizon is segmented into two parts: a short first prediction horizon for real-time control optimization, and a long second prediction horizon for pre-computing safe terminal states. This segmentation allows the system to benefit from long-horizon safety guarantees without incurring the full computational cost during real-time control.
Solution Approach 2:
Safe terminal states at the long prediction horizon are pre-computed and stored in a lookup table before real-time control is needed. During real-time operation, the system only needs to query this pre-computed data, avoiding the computational burden of long-horizon optimization while maintaining safety guarantees.
2Ease of operation
If predictive optimization is constrained to a short prediction horizon to reduce computational complexity, then the ease of operation is improved, but the risk of unfeasible vehicle states increases
Solution Approach 1:
The optimization process uses feedback from the pre-computed safe terminal states at the long horizon to guide the short-horizon control decisions. By incorporating this feedback through the cost function and constraints, the system ensures that short-horizon actions are directed toward feasible and safe terminal states, eliminating the risk of unfeasible vehicle states.
Solution Approach 2:
The pre-computed safe terminal states act as an intermediary between the short-horizon controller and the long-horizon safety requirements. This intermediary provides the short-horizon optimizer with information about safe destination states, enabling it to make computationally efficient decisions that guarantee feasibility and safety.
3Productivity
If a short prediction horizon is used to maintain real-time control capability, then the productivity is improved, but the measurement precision of long-term vehicle behavior prediction deteriorates
Solution Approach 1:
Long-horizon safe terminal states are pre-computed and stored before real-time control execution. This preliminary action allows the system to have accurate long-term prediction information available instantly during real-time control, eliminating the need to perform computationally intensive long-horizon predictions during operation.
Solution Approach 2:
Instead of performing real-time long-horizon optimization, the system uses pre-computed safe terminal states as copies or representatives of long-term behavior. These copied states provide the necessary prediction accuracy for safety-critical decisions without requiring actual long-horizon simulation during real-time control.
Data Source
AI summary
A computer system including processing circuitry configured to: determine a set of vehicle control parameters, using predictive optimization of a vehicle model with a first prediction horizon, the predictive optimization being constrained by a set of allowed vehicle states at the first prediction horizon; and control an operation of a vehicle using the determined set of vehicle control parameters, wherein: the set of allowed vehicle states at the first prediction horizon is a set of initial vehicle states estimated to result in a set of safe vehicle states at a second prediction horizon longer than the first prediction horizon.