Learning controller for vehicle control
A learning controller overcomes tuning problems in vehicle simulation programs by estimating requisite vehicle-specific parameters, effectively learning from its mistakes, as the vehicle is automatically driven around a track. After a sufficient period of calibration, the learned parameters are automatically saved to a car-specific file. The file parameters may be loaded in the controller in the future to optimally control a vehicle without the need to re-run the learning procedure.
1. A method for optimizing a set of vehicle-dependent control parameters for use in controlling a vehicle, the method comprising
establishing an initial set of vehicle-dependent control parameters, the vehicle-dependent control parameters comprising car-specific values for control functions including at least one of braking, accelerating or steering;
computing a first set of vehicle control values based upon the initial set of vehicle-dependent control parameters;
computing a first set of motional data for the vehicle at a first time;
controlling movement of the vehicle based upon the first set of vehicle control values;
computing a second set of motional data for the vehicle at a second time later than the first time;
computing a revised set of vehicle-dependent control parameters based upon a comparison of the second set of motional data with the first set of motional data;
computing a second set of vehicle control values based upon the revised set of vehicle-dependent control parameters;
without user intervention, iterating through these operations by replacing the initial set of vehicle-dependent control parameters in the establishing operation with the revised set of vehicle-dependent control parameters for each successive iteration until an optimal set of vehicle-dependent control parameters is computed;
storing the optimal set of vehicle-dependent control parameters; and
using the stored optimal set of vehicle-dependent control parameters by a control system to maintain a computer-controlled car along a path in a computer video game.
2. The method of claim 1 , wherein the revised set of vehicle-dependent control parameters is dependent upon a velocity of the vehicle.
3. The method of claim 1 further comprising
subdividing the revised set of vehicle-dependent control parameters into discrete value ranges; and
storing the revised set of vehicle-dependent control parameters in a single-dimension table according to the discrete value ranges.
4. The method of claim 1 further comprising
subdividing the revised set of vehicle-dependent control parameters into discrete value ranges; and
storing the revised set of vehicle-dependent control parameters in a multi-dimension table according to the discrete value ranges.
5. The method of claim 1 further comprising adapting the second set of vehicle control values to control at least one of vehicle steering, vehicle acceleration, and vehicle braking.
6. A method for optimizing a set of vehicle-dependent control parameters for use in controlling a vehicle, the method comprising
during vehicle simulation, inputting an initial set of vehicle-dependent control parameters into a controller;
computing under direction of the controller an initial set of vehicle control values based upon the initial set of vehicle-dependent control parameters;
inputting the initial set of vehicle control values into a physics engine;
controlling movement of the vehicle under direction of the controller based upon the initial set of vehicle control values;
inputting a first set of motional data for the vehicle at a first time output by the physics engine into a learning system;
inputting a second set of motional data for the vehicle at a second time later than the first time output by the physics engine into the learning system; and
computing a revised set of vehicle-dependent control parameters under direction of the learning system based upon a comparison of the second set of motional data with the first set of motional data;
computing a second set of vehicle control values under direction of the controller based upon the revised set of vehicle-dependent control parameters;
without user intervention, iterating through these operations by replacing the initial set of vehicle-dependent control parameters in the first inputting operation with the revised set of vehicle-dependent control parameters for each successive iteration until an optimal set of vehicle-dependent control parameters is computed;
storing the optimal set of vehicle-dependent control parameters; and
using the stored optimal set of vehicle-dependent control parameters by a control system to maintain a computer-controlled car along a path in a computer video game.
7. The method of claim 6 , wherein the revised set of vehicle-dependent control parameters is dependent upon a velocity of the vehicle.
8. The method of claim 6 further comprising
subdividing the revised set of vehicle-dependent control parameters into discrete value ranges under direction of the learning system; and
storing the revised set of vehicle-dependent control parameters in a single-dimension table under direction of the learning system according to the discrete value ranges.
9. The method of claim 6 further comprising
subdividing the revised set of vehicle-dependent control parameters into discrete value ranges under direction of the learning system; and
storing the revised set of vehicle-dependent control parameters in a multi-dimension table under direction of the learning system according to the discrete value ranges.
10. The method of claim 6 further comprising adapting the second set of vehicle control values under direction of the controller to control at least one of vehicle steering, vehicle acceleration, and vehicle braking.
11. A method for optimizing steering parameters for using in controlling a vehicle, the method comprising:
(a) using a processing unit coupled to a memory, initializing running average variables stored in a two-dimensional table indexed based on a steering angle and a velocity;
(b) computing a current vehicle heading and a current vehicle velocity;
(c) accessing a racing line, computing a target point and calculating a desired yaw rate;
(d) running a control process which produces a desired steering angle;
(e) running one game physical simulation step for a pre-specified time period;
(f) computing a new car heading;
(g) computing an actual yaw rate over the pre-determined specified time period;
(h) computing table indices based on the steering angle and the velocity;
(i) updating running average variables of an entry stored in the two-dimensional table indexed based on computed tables indices;
(j) iterating through steps (b)-(i) until a pre-determined criteria has been met; and
(k) saving the running average variables stored in the two-dimension table to a vehicle-specific file.
12. The method of claim 11 , wherein the pre-determined criteria is based on a measure of accuracy of the control process.