Improving Vehicle Comfort and Safety Using Digital Human Modeling

Transcription

1 Improving Vehicle Comfort and Safety Using Digital Human Modeling Matthew P. Reed, PhD Research Associate Professor Head, Biosciences Group, UMTRI Director, Human Motion Simulation Lab, Center for Ergonomics, Industrial and Operations Engineering

2 University of Michigan Chrysler HQ Nissan Tech Center General Motors Tech Center Toyota Tech Center General Motors HQ Hyundai-Kia Tech Center Ford HQ and Research Labs

3 U-M Transportation Research Institute Behavioral Sciences Group Biosciences Group Driver Interface Group Engineering Systems Group Human Factors Group Vehicle Safety Analytics Group Vehicle Systems and Control Group Young Driver Behavior and Injury Prevention Group Center for the Management of Information for Safe and Sustainable Transportation (CMISST) ~150 research faculty and staff, $25M annual research volume

4 Center for Ergonomics Located in Industrial and Operations Engineering Led by Prof. Tom Armstrong and 11 other faculty researchers Human Motion Simulation Laboratory founded by Prof. Don Chaffin and Prof. Julian Faraway in 1998

5 Lab Overview Biosciences Group Engineering Anthropometry and Ergonomics" Laboratory and Field Ergonomics Studies" Human Motion Simulation Lab Injury Biomechanics" Motion Modeling" " Crash Data Analysis" Advanced Manikin Development"

6 Humans in Engineered Systems Humans interact with nearly all products and systems either in manufacture, use, maintenance, or disposal. On nearly every relevant attribute, humans vary more than any other factor considered in the design of a product Human physical variability (size, shape, mass, strength ) is a critical consideration in the design of products to be used by people

7 Overview Posture Prediction and Vehicle Occupant Accommodation" Body Shape Modeling and Advanced Manikin Development" Human Motion Simulation" Major Challenges in DHM"

8 Posture Prediction and Accommodation

9 Posture Prediction and Accommodation Goals: 1. Predict the postures of individual vehicle occupants for human figure model analyses 2. Predict the distribution of occupant locations for accommodation Methods: 1. Body landmark measurement using FARO Arm in vehicle and laboratory conditions 2. Statistical modeling of body landmark locations

10 What is Anthropometry? Human Body + Spatial Measurement = human size and shape Often used to mean anthropometric data or the application of anthropometric data Anthropometry is critical in the design of most engineered systems with which people interact Anthropometry is the oldest and most successful component of human factors engineering Traditional Anthropometric Dimensions"

11 Functional Anthropometry Standard Anthro for Target Population (NHANES, etc.)" SAE Standards: J941, J1052, J4002, J4003, J4004" Vehicle and Seat Geometry" Statistical Analysis" Posture prediction models for human figure models (Jack etc.)" Measure Landmark Locations" Seat design dimensions and tools"

12 Posture Prediction: Major Models Cascade Posture Prediction Passenger Cars and Light Trucks Implementation: Jack (OPT), RAMSIS, Catia (VOA) Large Trucks and Buses Implementation: Jack (OPT), Catia (VOA)

13 Accommodation: Major Models SAE J1052: Driver Head Contour SAE J941: Driver Eye Location SAE J4004: Seating Accommodation

14 Multivariate Functional Anthro Vehicle and Seat Geometry" Target Body Dimensions: Stature, Body Weight, " Regression" Measure Landmarks in Seated Postures" Principal Component Analysis" Whole-Body Landmark & Joint Configurations" Joint Center Location Estimates" Large Child Omnidirectional Dummy (Humanetics ATD) Crash Dummy Anthro Specs"

15 Human Motion Simulation

16 Research Approach 1. Overall goal is to improve digital human modeling software so that people can perform faster, more accurate ergonomics analyses Focus: Vehicle assembly & vehicle occupant interactions 2. Detailed study of human motions in the laboratory 3. Focus on people performing generic tasks of interest: reaching, moving objects, stepping 4. Modeling of motions: goal is accurate simulation of new motions that have not been measured in the lab 5. Integrated consideration of biomechanics and ergonomics limits

17 Effects of Hand Force on Posture Suzanne Hoffman Ph.D. " Prediction of posture in standing tasks based on biomechanical principles"

18 Transit Model (David Wagner, PhD) Transition Stepping and Timing 1. Select stepping behavior based on task and operator characteristics 2. Scale transitions steps 3. Integrate with gait strides

19 Reference Implementation in Jack The HUMOSIM Framework is a set of statistical models, kinematic control algorithms, and biomechanical prediction methods. The demonstration and development implementation of the HUMOSIM Framework is written in Python for execution in the Jack human modeling system from UGS. The Reference Implementation uses only graphics and forward-kinematics functionality from Jack -- all other functionality is based on HUMOSIM algorithms

20 HUMOSIM Framework Task Input: Independent of Current State of Environment and Figure 1. Place object A on object B 2. Use tool C with object A 3. Figure and Environment Information: Figure posture and position, object position and mass, obstacles, Task Planner Behavior Selection Mid-Level Task Plan: Walk, pick up, carry, place, manipulate, etc. What to do?

21 HUMOSIM Framework Motion Planner How to do it? Transit Model: foot placement and timing Grasp and Object Manipulation Planner: Hand placement and grasp Resource-Based Component Scheduling Model: sequence and timing of head, upper-extremity, and torso motions Biomechanical Motion Planner: posture/motion targets based on biomechanical criteria Component-Level Motion Plan: Steps, reaches, grasps, manipulations, gaze transitions, Current Reference Implementation in Jack

22 HUMOSIM Framework Component-Level Motion Plan: Steps, reaches, grasps, manipulations, gaze transitions, hmshuman Motion Modules Gaze Module: head and eye control Task Element Coordinator Schedule motor elements, plan sub-elements (foot movements, compensatory torso motions) Upper-Extremity Modules: hand position and orientation Hand Module: grasp and object manipulation Simulation Clock Task Element Dispatcher All modules incorporate feedback based on monitoring of current figure state relative to goals Torso Module: pelvis orientation and lumbar motion Pelvis Module control pelvis trajectory and balance Right Lower-Extremity Module: step following Left Lower-Extremity Module: step following Planned Current

23 Framework as Motion Parameterization The Framework generates motions as transitions between endeffector targets Gaze/hand/foot targeting is a useful midlevel description of many tasks Behavior-based obstacle avoidance adds non-contact interactions Preliminary ingress results generated using foot, hand, gaze, and pelvis targets Framework automatically adjusts for different figure sizes, seat heights, seat positions, vehicle heights, roof positions, and sill positions

24 Effects of Hand Force: Simulation One Hand, No Force, Fixed Feet One Hand, Variable Force, Fixed Feet Two Hands, No Force, Fixed Feet Two Hands, Variable Force, Moving Feet

25 More on Automatic Posturing Low Force, Auto Foot Placement Low Force, Obstacles

26 Gait and Transition Stepping

27 Five-Phase Hand Posture During Reach Reaching includes five phases: Release Grasp Relax Transport Open

28 Simulated Grasp Motions Based on a library of hand motions gathered with CyberGlove Two-Finger Pinch Precision Pinch Power Grasp Lateral Pinch Chuck Pinch

29 Grasp Model Components Movement Planning Runtime Grasp location planning, including hand end effector location Hand trajectory execution (6 dof) Grasp pose trajectory planning (PCA model) Hand pose trajectory execution (22 dof) Reach hand trajectory planning Collision-based pose tuning

30 Integration with Framework Analyze bounding box of grasp object! select grasp type and location Compute scaled aperture (object dimension/hand length) Compute hand pose motion for desired duration of open+grasp motion phases Place hand end effector site at computed location based on grasp type and object dimensions As reach is executed, use collision-based grasp to tune hand posture during open+grasp phases

31 Integration with Framework Sequential reach-grasps using automatically generated and manually positioned grasp targets Hand end effector is automatically repositioned with respect to hand based on planned grasp

32 HUMOSIM Framework in Action Toyota co-op and HUMOSIM student Helinda Ho conducted several case studies to test HUMOSIM Framework functionality.

33 HUMOSIM Framework in Action Comparisons of Framework (Jack) and Safework simulations by Toyota interns Helinda Ho and Rick Wetherhold

34 HUMOSIM Framework in Action Wei Zhou (PhD Student): Using Framework with a GUI developed by GM.

35 Driver Simulation Toolkit Wraps Reference Implementation functionality in high-level commands: steer right 45 degrees, shift from position 1 to position 2, press accelerator pedal Jack environment file contains typical targets: radio knob, forward vision target, left mirror vision target Designed to interface with QN- MHP in Virtual Driver via TCP/IP

36 Cognitive Model Connection The Virtual Driver: Linking the QN- MHP Cognitive Human Model with the HUMOSIM Framework Task Duration (s) 50" 40" 30" 20" 10" 0" Normal Vehicle Heavy Vehicle Short" Female" Midsize" Female" Midsize" Male" Tall"Male"

37 Implementation in Commercial Software We work closely with our technology partners to move our research findings from the laboratory to industry Siemens has implemented a wide range of Humosim models and algorithms into Jack v7. Dassault Systems has implemented several models into Delmia V6

38 Body Shape Modeling

39 Surface Anthropometry Methods 3D whole-body laser scanning has transformed anthropometry research In addition to traditional measures obtained with calipers and tape measures, we now record the whole body shape in various postures UMTRI s Whole-Body Laser Scanner"

40 UMTRI Lab Methods: Body Shape standard anthropometry whole-body laser scanner optical landmarks additional landmarks with FARO Arm multiple standing and seated postures Laser towers (4) Hand-held laser scanner to augment towers Red laser beam VITUS XXL Scanner

41 UMTRI Lab Methods: Body Shape Scanning requires 12 seconds, plus an additional minute for hand scanning in some postures Seated postures span a wide range of recline and lumbar spine flexion

42 Body Shape Modeling Whole-Body Scan Data Handheld Scanner Data Clean and Fit Polygon Mesh Manual Landmark Extraction Manually Measured Body Landmarks Model Integration Mesh with Landmarks Fit Homologous Mesh Standard Anthropometry PCA+Regression Analysis Statistical Model to Predict Body Shape from Standard Anthro or Landmark Locations

43 3D Anthropometry CAESAR Study: First large-scale 3D anthro survey of US civilians ( ). Scans of 2400 adults (18-65 years) in 3 postures. Statistical Approach: generate a parametric model of body shape as a function of predictors (gender, stature, weight, etc.) Cyberware Whole-Body Scanner"

44 3D Anthro: Statistical Modeling Statistical models of standing body shape using U.S. CAESAR data Parametric Standing Whole-Body Model

45 3D Anthro: Statistical Modeling

46 3D Anthro: Statistical Modeling Varying Stature, BMI, and Chest/Waist/Hip Ratios" Based on 1000 men and 1000 women from U.S. CAESAR"

47 Parametric Torso Models Varying Stature, BMI, and Posture" Application to Office Chair Design"

48 Application: Seat Fit Finite-element modeling of seat-back interaction for auto seat design: use a population of realistic 3D models rather than a few occupant sizes

49 Skeletal Anthropometry Parametric Modeling of Skeletal Structures 2 nd PC of Ribcage Shape CT Images Statistical Modeling Data Extraction (N > 100) Predicted Skeleton Size and Shape = ƒ (stature, mass, ) Hybrid-III vs. Human Pelvis

50 Measuring Children UMTRI has gathered 3D anthropometry for 160 children ages 4-11 for crash safety applications Raw scan data for some of the study postures"

51 Measuring Children 3D models for some of the study postures"

52 ATD (Crash Dummy) vs Child

53 Elderly Posture, Body Shape, and Belt Fit Age 20 yrs Age 65 yrs Statistical models of body shape: effects of age (Example from UMTRI analysis of U.S. CAESAR Data) Older Adults (N=180) Application to ATDs and Human Body Models

54 Elderly Posture, Body Shape, and Belt Fit 21 years 95 years

55 Anthropometry for Test Devices? Standard anthropometric data are gathered in standing and erect seated postures with minimal clothing Seated posture and gear affect body shape and space requirements

56 UMTRI Seated Soldier Posture Ongoing Army-funded study to quantify seated soldier posture and body shape, including effects of body-borne equipment Body landmark locations and whole-body shape measured in a range of standing, driver, and crew seat conditions

57 Effects of Equipment on Space Claim Overlaid scans without posture adjustment

58 Applications of 3D Anthro Seat Design" Dummy Development" Human Body Models for Injury Prediction" Figure Models for Ergonomics" Vehicle Interior Layout (Packaging)"

59 Major Challenges in DHM

60 Major Challenges in DHM 1. Representing variability (size, shape, behavior, capability ), not just a few examples Problem Manikins don t look like people Solution Improved parametric body shape, posture, and motion modeling Takes a long time to do a simulation Automated simulation tools Summarizing results of multiple simulations Improved statistical tools with graphical presentation

61 Major Challenges in DHM 2. Accurate, repeatable, reproducible analyses; result is not dependent on the expertise of the software user Problem No data available for most tasks Manual intervention by software users causes problems with repeatability Different software users get different answers Solution Increased laboratory and field research and better representation of results Automated simulation tools Better training and reduced user input (less need for expertise)

62 Major Challenges in DHM 3. Powerful, flexible tools for use by engineers without formal training in ergonomics Problem Software is hard to use Software results requires subjective interpretation Better integration with CAD and engineering analysis tools Solution Automated simulation with only high-level task description Software, rather than user, should do interpretation and prepare report Stand-alone software requires additional training and licenses rarely used

63 Acknowledgements This research has been supported in part by Chrysler Faurecia Ford Motor Company General Motors Herman Miller Navistar U.S. National Highway Traffic Safety Administration U.S. National Institute for Occupational Safety and Health Toyota Technical Center U.S. Army mreed.umtri.umich.edu