A Robotic Wheelchair for Crowded Public Environments

1
A Robotic Wheelchair for Crowded Public
Environments
EE887 Special Topics in Robotics Paper Review

• 2001. 6. 7.
• Choi Jung-Yi

E. Prassler, J. Scholz, and P. Fiorini, A
robotic wheelchair for crowded public
environments, IEEE Robotics Automation
Magazine, vol. 7, no. 1, pp. 38-45, 2001
2
Overview

• Two difficult situations of using wheelchair
• Form conversations with the user community
• Navigation in
• NARROW CLUTTERED environments
• WIDE CROWDED areas
• MAid (Mobility Aid for Elderly and Disabled
  People) Combines
• Narrow Area Navigation (NAN) Behavior
• ? Semiautonomous Navigation Mode
• Wide Area Navigation (WAN) Behavior
• ? Autonomous Navigation Mode

3
Hardware Design

• Mechanical Part
• Rear wheels two differentially driven
• Front wheels two passive castor
• Maximum speed 6 km/h (Powered by 12 V battery)

4
Hardware Design

• Central Processing
• Industrial PC(Pentium 166 MHz) QNX
• Sensors
• Dead-reckoning system wheel encoders optical
  fiber gyroscope
• 3 x 8 Ultrasound transducers and microcontrollers
• Short-range sensing two infrared scanners
• 2-D laser range-finder

5
Hardware Design (Contd)
6
Control Architecture

• WAN Hierarchical Control Architecture

Strategic Level
Tactical Level
Basic Control Level
7
Basic Control Level
8
Tactical Level (Overview)

• The core of WAN Module
• Motion Detection
• Motion Tracking Obstacle Velocity Estimation
• Computation of the Evasive Maneuvers

9
Tactical Level (Overview) contd

• Monitoring the surrounding environment
• Detect the environment objects
• Identify stationary / moving object

• Sonar system

Laser range finder

• Estimate the speed and direction of the object

Past trajectory and velocity

• Determine if MAid is moving on s collision course
  with objects
• Compute the avoidance maneuver

10
Strategic Level

• Main task
• Navigating in crowded area
• Reaching a specific goal
• Without any intermediate goal
• Selection the nest motion goal by the user
• Strategic level will be expended by including a
  path planner capable of adding the computation of
  subgoal sequences

11
Motion Detection and Tracking

• A sequence of single observation
• Investigating where these observations differ
  from each other
• Discrepancy ? potential change
• Occupancy Grid Representation
• A projection of the range data on a 2-D
  rectangular grid
• Grid element ? a small region of the real world
• Updating every cell ? time consuming process

12
Time Stamp Map

• Modification of occupancy grid representation
• Map only cells observed as occupied
• Cell coinciding with the range measurement
• All other cells ? left untouched
• Range image ? 200 x 200 time stamp map
• Takes 1.5 msec on a Pentium 166 MHz

13
Motion Detection Algorithm

• Based on a simple heuristic
• Cell is occupied
• by a stationary object if corresponding cells in
  TSMt and TSMt-1 carry time stamps.
• By a moving object if corresponding cells in TSMt
  carry a time stamp different from TSMt-1 or no no
  time stamp at all.
• TSMt Time Stamp Map at time t

14
Motion History

• Objects are represented by cell ensembles in the
  sensor map.
• Identifying the object in a sequence of maps
• Correspondence between objects
• ? using a nearest-neighbor criterion based on a
  Euclidean distance
• The ensembles describes the same object
• ? if the distance to the nearest neighbor is
  smaller than a certain threshold.
• Threshold
• For stationary object 30cm
• For moving object 1 m

15
Motion Planning

• For simplicity
• Model the wheelchair and the obstacles as
  circles.
• ? Planar problem with no rotations

obstacle
Wheelchair
16
Velocity Obstacle

• VO of A with respect to B
• Identifying the set of velocities of A causing a
  collision with the obstacle B at some time
• To avoid collision selecting the tip of VA
  outside VO

17
Velocity Obstacle (contd)

• Collision Cone v.s. Velocity Obstacle
• Avoiding multiple obstacles
• Prioritization among Vos

18
Velocity Obstacle (contd)

• Consideration of wheelchair dynamics
• Some heuristics for making trajectory

Reachable Avoidance Velocity
Velocity Obstacle
Reachable Velocity
Toward Goal
Maximum Velocity
Structure
19
Experiments in Real Situations

• Roaming in a Railway Station
• Hall size 20 x 40 m2
• Several tens of people
• Survived about 18 hours
• Hannover Fair 98
• Survived more than 36 hours