Visible to the public File preview

Active Safety Control in  A ti S f t C t l i Automotive Cyber‐Physical Systems Automotive Cyber‐
PI: Francesco Borrelli PI: Francesco Borrelli
Email: fborrelli@me.berkeley.edu Department of Mechanical Engineering  University of California y f f Berkeley, USA www.mpc.berkeley.edu

Co‐PI Karl Hedrick,  Co‐PI Karl Hedrick Ruzena Bajcsy PI Karl Hedrick, Ruzena

Automotive Cyber‐Physical System  Automotive Cyber‐
Vehicle  Actuators

Vehicle and Tire Sensor Data

Intelligence I t lli

Driver  Model/Intent

Safety

Comfort

Efficiency

Lane Departure A14 Highway – Lane Departure A14 Highway – June 2009

www.mpc.berkeley.edu

2008 US Statistics

www.mpc.berkeley.edu

Automotive Cyber‐Physical System  Automotive Cyber‐
Vehicle  Actuators

Vehicle and Tire Sensor Data

Intelligence I t lli

Driver  Model/Intent

Safety

Comfort

Efficiency

CPS‐ CPS‐Synoptic Scheme Human Vehicle

Environment

CPS‐ CPS‐Synoptic Scheme Human Vehicle

Environment

CPS‐ CPS‐Synoptic Scheme Human Vehicle

Environment

Human‐Vehicle Interaction
Steering Angle

Actuation  System

Wheels Traction Torque Wheels  Braking Torques ……..

Human‐Vehicle Interaction
Steering Angle

Actuation  System

Wheels Traction Torque Wheels  Braking Torques ……..

Hydraulic Brake Unit

CPS‐ CPS‐Synoptic Scheme Human Vehicle

Environment

Vehicle‐ Vehicle‐Road Interaction FEM Simulation

www.mpc.berkeley.edu

Vehicle‐ Vehicle‐Road Interaction Simplified Models
Longitudinal Force Dry Asphalt

Maximum Braking B ki

Maximum Acceleration A l i Steer Angle

Tire Slip Tire Slip
Longitudinal Force

Lateral Force

www.mpc.berkeley.edu

Vehicle‐ Vehicle‐Road Interaction Simplified Models
Longitudinal Force Dry Asphalt S o Snow

Maximum Braking B ki

Maximum Acceleration A l i Steer Angle

Tire Slip Tire Slip
Longitudinal Force

Lateral Force

www.mpc.berkeley.edu

Vehicle‐ Vehicle‐Road Interaction p Simplified Models
Longitudinal Force Dry Asphalt S o Snow Ice
Maximum Braking B ki Maximum Acceleration A l i Steer Angle

Tire Slip Tire Slip
Longitudinal Force

Lateral Force

www.mpc.berkeley.edu

CPS‐ CPS‐Synoptic Scheme Human Vehicle

Environment

Environment‐ Environment‐Human Interaction

Sensory Inputs

Brain

Spinal Cord

Muscles

http://www.nature.com/nrn/journal/v5/

CPS‐ CPS‐Synoptic Scheme Human
Vehicle Model  Environment  E i t Model 

Vehicle

Environment
``We know that a lot of the brain has an internal neural simulator”…   We know that a lot of the brain has an internal neural simulator “to anticipate or predict the future for a given a input” Eric Kandel (Charlie Rose interview, 2008) 

CPS‐ CPS‐Synoptic Scheme Human
Vehicle Model  Environment  E i t Model 

Vehicle

Environment

CPS with Driver Assistance System (DAS) Human
Vehicle Model  Environment  E i t Model 

Vehicle DAS

Environment

Vast Majority of DAS systems Human
Vehicle Model  Environment  E i t Model 

Vehicle DAS

Environment

AntiAnti-lock Braking Systems
Longitudinal Force

Maximum Braking g

Maximum Acceleration Steer Angle

Long. Slip g p
Longitudinal Force

Lateral Force

CounterCounter-Steering and Over-Steering Over-

Vast Majority of DAS systems Human
Vehicle Model  Environment  E i t Model 

Vehicle DAS

Environment

Vehicle CPS Main Issues
• • • • Complexity/Compositionality… y y “Shy” Autonomy No Guarantees/Heuristic Tunings u a o o e cep o / og o g o ed Human MotionPerception/Cognition Ignored 

CPS Main Issues
• • • • Complexity/Compositionality y y “Shy” Autonomy No Guarantees/Heuristic Tunings u a o o /pe cep o /cog o g o ed Human Motion/perception/cognition ignored 

Research Goals
Provably Safe and Adaptive Autonomy 
Quantifying Uncertainties (Human and Environment)

Quantifying Uncertainties:  Tire/Road Interaction Tire/Road Interaction

CYBER™TYRE

Quantifying Uncertainties:  Tire/Road Interaction Tire/Road Interaction
Tangential and Lateral Sidewall Tangential and Lateral Sidewall Tange ent

Lateral

Quantifying Uncertainties: Driver Models 
Objective: Exploit real‐time driver state (joint angles)  detection to extract compatible set of trajectories

Methodology:  1‐Stereo cameras for 3D reconstruction of the scene 2‐ Articulated tracking, use body sensor network for validation

Real‐ Real‐Time 3D Reconstruction and the  Middlebury Stereovision Dataset Middlebury Stereovision Dataset

Real‐ Real‐Time 3d Reconstruction Driver

Provably Safe Adaptive Autonomy
Vehicle Actuators Vehicle and Tire Sensor Data

Adaptive and Predictive  p Autonomy

Driver Intent

www.mpc.berkeley.edu

Set‐ Set‐Based Control Design
• Compute N‐steps controllable set Compute N steps controllable set



• Given      compute      such that 



Set‐ Set‐Based Control Design
• Compute N‐steps controllable set Compute N steps controllable set



Benefit in Autonomy Concept
At time j • Driver intent  • Among all possible actuations        choose the one that  solves 

Adaptive and Predictive Autonomy p y
www.mpc.berkeley.edu

Main Limitation: Real‐Time Computation Main Limitation: Real‐ ``Predictive Control”: Borrelli, Bemporad Morari
• PWA Model
www.mpc.berkeley.edu www mpc berkeley edu

• One‐step robust controllable set
‐For linear (A,B) system: ‐For linear systems, result is polytope For linear systems result is polytope ‐For PWA systems, result is union of polytopes

Process Industry 4Ghz, 4Ghz 1 Terabyte

Automotive 50Mhz, Mbytes 50Mh 2 Mb t

Safe Control Design Through  Simplified Models Simplified Models
“True” System  Simplified System 

……
XN −2 ZN−2

XN −1 ZN −1

:   k‐step controllable set for “True” System when control law  p y designed for simplified system is applied.

2D Example
States: Inputs: Disturbance: Assume constant: A
  y ,

M

f  ,Vx

PWA approximation of Pacejka tire model PWA approximation of Pacejka tire model

2D Example – Robust Set Computation 2D Example –
0.6 0.4 0.2

0 x2 -0.2 -0.4 -0.6 -0.8 -2

0.6 0.4 0.2 0 x2 -0.2 -0.4 -0.6 -0.8 270 272 274 276 x1 278

-1.5

-1

-0.5

0 x1

0.5

1

1.5

2

280

282

From 2‐D to 12‐ From 2‐D to 12‐D Example Experimental results @ 72 Kph on Ice Experimental results @ 72 Kph on Ice Experimental results @ 72 

Conclusions/Outlook
• Developing concept and methods for Provably Safe Adaptive Autonomy – Quantifying uncertainties in  • Vehicle/Road Interaction  • Human/Car Interaction – Real‐time Computation of Controllable Sets  p with different level of granularity

The Team
UC Berkeley 
• • • • • • • • • Gurkan Erdogan Ramanarayan Vasudevan Ricardo Cervera Ricardo Cervera Navarro Sanghyun Hong Theresa Lin Ye Zhuang Yiqi Gao Ruzena Bajscy Karl Hedrick

Ford Research Labs (Dearborn,USA) • Jahan Asgari, Eric Tseng, Davor Hrovat Pirelli Research Labs (Milano,Italy) ( , y) • Federico Mancosu, Giorgio Audisio