Identification of linear and nonlinear driver steering control David Cole, Andrew Odhams, Steven Keen Vehicle Dynamics and Control 211 Fitzwilliam College, Cambridge, 5 April www.dynamics.org
Driver-Vehicle Dynamics Group Aim: measure, understand and model driver- behaviour shift engineering activity from prototype phase to design phase reduce cost and improve performance
Driver- model structure handwheel torque target path controller (brain) neural signals neuromuscular system handwheel angle dynamics response learning internal model (brain) + sensory system sensory feedback
target path Simpler model structure for this study model predictive control controller (brain) low-pass filter neuromuscular system handwheel angle constant speed dynamics response internal model (brain) nonlinear tyres How well does this model replicate measured steering behaviour?
Model predictive control target path
Model predictive control target path handwheel angle sequence 1 2 3 4 prediction horizon
Model predictive control target path handwheel angle sequence 1 2 3 4 prediction horizon predicted path
J = Cost function, weighting values N! q y y error j j=1 ( ) 2 + q!! error ( j) 2 +! ( j) 2 q y y error!!!! error! q!
Experiment 1: driving simulator linear lateral-yaw model randomly curved path, 4km long 5 test subjects, various experience constant speed between 2m/s and 4m/s
Identification procedure q y, q! driver model simulated handwheel angle target path response driving simulator test driver + + + error measured handwheel angle weighted error weighting filter noise
.5.4.3.2.1. Identification results q y q! driver 1 driver 2 driver 3 2 25 3 35 4 speed / m/s 3 2 1 2 25 3 35 4 speed / m/s
Experiment 2: test, double lane change ISO double lane change, <4m/s 2, 14 drivers
handwheel angle / deg δ ( ) δ ( ) handwheel angle / deg Measured handwheel angle histories two drivers 5 5 1 12 14 16 18 2 5 Driver 3 mean± Driver 5 5 1 12 14 16 18 2 distance / m
Identification procedure indirect method q y, q! target path direct indirect response driver model model test test driver + + + simulated handwheel angle error measured handwheel angle weighted error weighting filter noise
Measured and simulated handwheel angle histories handwheel angle / rad 1-1 Driver 3 1 15 2 distance / m handwheel angle / rad 1-1 Driver 5 simulated average measured 1 15 2 distance / m
Conclusions: linear regime driving simulator and test track measurements noise model identified; used to reduce bias model predictive controller fits measurements well cost function weights represent inter-driver variation
Experiment 3: test, ISO elk avoidance, ~9m/s 2 Less-experienced drivers evidence of learning action handwheel δf δfa angle / rad 4 2 2 4 Trials 1-3 8(1) Trials 4-6 8(2) Trials 7-9 8(3) 15 2 4 2 2 4 15 2 distance / m 4 2 2 4 15 2 individual trial mean of 3 trials
Experienced drivers: inter-driver variation, but no learning 4 1 9 4 δfa handwheel angle / rad δf δfa δf 2 2 4 4 2 2 2 2 4 14 16 18 2 14 16 18 2 13 14 4 2 2 4 14 16 18 2 x GC distance / m 4 14 16 18 2 x GC
Model predictive control, nonlinear target path handwheel angle sequence 1 2 3 4 lateral force slip angle
Internal model range lateral tyre force novice driver linear ( =) mid-range ( =.5) full-range ( =1.) expert driver tyre slip angle
Measured and simulated handwheel angle histories handwheel angle / rad 2-2 Driver 1 14 16 18 2 distance / m =.5 handwheel angle / rad 2-2 Driver 13 simulated avg. measured 14 16 18 2 distance / m =1.
Conclusions: nonlinear regime model predictive control extended to nonlinear regime expert drivers simulated well identified internal model range ( ) not as expected inexperienced drivers show learning behaviour
Further work combined steering and speed control control strategies for limit manoeuvres learning strategies co-operative control sensory dynamics; perception thresholds
Bibliography For a full list see www.dynamics.org Odhams AMC, Identification of driver steering and speed control, PhD thesis, University of Cambridge, 26. Odhams AMC and Cole DJ, Application of linear preview control to modelling human steering control, Proc. IMechE, Part D: J. Automobile Engineering, 29, 223 (D7), 835-853. Odhams AMC and Cole DJ, Identification of a driver s preview steering control behaviour using data from a driving simulator and a randomly curved road path, Proc. AVEC 21, Loughborough, August 21. Keen SD, Modeling driver steering behavior using multiple-model predictive control, PhD thesis, University of Cambridge, 28. Keen SD and Cole DJ, Application of time-variant predictive control to modelling driver steering skill, Vehicle System Dynamics, 7 December 21 (ifirst). Keen SD and Cole DJ, Bias-free identification of a linear model-predictive steering controller from measured driver steering behavior, submitted to IEEE Transactions on Systems, Man, and Cybernetics.