Con trol Algorithm

Investigators have taken a variety of approaches to implement control software for smart wheelchairs, based on the functions supported by the smart wheelchair and the sensors it uses. Many smart wheelchairs implement semiautonomous behaviors (combining or switching between input from the driver and the intelligent system) rather than autonomous driving behaviors (where the driver’s input is ignored) to offer the driver higher levels of control while ensuring safety. However, the exact mechanism of semiautonomous control that is both effective and acceptable to users is still under investigation (Viswanathan, Zambalde et al. 2016).

Critical Review of the Technology Available

Smart wheelchairs are at level 4 of NASA’s progression of technology readiness levels (Mankins 1995); low-fidelity systems have been used to demonstrate basic functionality in controlled environments. Although a large number of prototypes have been described in the research literature, no smart wheelchairs are widely used today. The CALL Center smart wheelchair is sold in the United Kingdom and Europe by Smile Rehab Limited (Berkshire, U.K.) as the “Smart Wheelchair” (http://smilesmart-tech.com /assistive-technology-products/smile-smart-wheelchair.php), but this is intended to be a training tool rather than a permanent mobility aid. An overview of smart wheelchair designs described in the research literature can be found in Table 5.4.

A significant technical obstacle to commercialization is the cost/accu- racy trade-off that must be made with existing sensors. There is reason for optimism, however, that the significant research in self-driving automobiles (Guizzo 2011) will lead to affordable technology that can also be used in smart wheelchairs. In 2014, the technology used in one of Google’s self-driving cars cost approximately $320,000 (Tannert 2014),

TABLE 5.3 Summary of Feedback Modalities Used by Smart Wheelchairs

Modality

Implementation

Advantages

Disadvantages

Auditory

Feedback can be provided through intermittent sound (e.g., warning sirens), continuous sound (e.g., tone indicating distance to the nearest obstacle), or spoken words.

  • • Does not distract visual attention
  • • Low cost
  • • May be washed out by ambient noise
  • • Nonspoken feedback may be misinterpreted
  • • Only available for a short period of time

Visual

Feedback can be provided through something as simple as individual lights or as complex as a “dashboard” implemented on a computer screen.

  • • Not affected by ambient noise
  • • Remains available to user
  • • Low cost

• Can distract visual attention

Haptic

Feedback can be provided through pressure applied to a joystick.

  • • Possible to give very precise feedback about the environment around the PWC
  • • Does not distract visual attention
  • • Only works if the driver uses a joystick to operate the PWC
  • • Expensive

TABLE 5.4 Overview of Semiautonomous Smart Wheelchairs

System

Functions

Control Method

Feedback

Sensors

Navigation

Feedback simulation (Wang, Gorski et al. 2011; Wang, Mihailidis et al. 2011)

Collision

avoidance

Teleoperator

Audio (direction) Haptic (joystick blocked in obstacle direction)

Visual (LEDs around joystick)

N/A

N/A

Wizard-of-Oz simulation (Mitchell et al. 2014; Rushton et al. 2014; Viswanathan, Zambalde et al. 2016)

Collision

avoidance

Teleoperator Three modes:

  • 1. Basic safety (stop to avoid collision, blocks direction of obstacle)
  • 2. Steering correction (turns away from obstacle)
  • 3. Autonomous

Audio (chair actions, directions) Haptic (direction or speed)

N/A

N/A

Anticollision contact sensor skirt (Wang, Gorski et al. 2011)

Collision

avoidance

Stops when bumper hits obstacle

Indicator lights (directions)

Microswitches

Lights indicate possible directions following stop through sensor skirt contact

(Continued)

System

Functions

Control Method

Feedback

Sensors

Navigation

Collaborative Assistive Robot for Mobility Enhancement (CARMEN) (Urdiales et al. 2010)

Collision

avoidance

Continuously shared control weighted based on local evaluation of human and robot efficiency

None

Laser range finder Encoders

Three layers: safeguard; reactive (potential fields approach [PFA]: user and wheelchair goals are attractors); deliberative (reach intermediate goals)

Efficiency based on smoothness (angle between current direction and upcoming motion vector); directness (angle between vector and destination); and safety

Collaborative controller (Carlson and Demiris 2012)

Obstacle

avoidance

Navigation

assistance

Shared dynamically

None

Camera (twodimensional [2-D] markers on ceiling)

Known indoor mapped environment; computer vision-based localization system using markers Determine user intent, verify action is safe with dynamic local obstacle avoidance (DLOA) algorithm If user deviates from safe minitrajectory, user is guided toward the minitrajectory with speed proportional to joystick deflection in that direction

TABLE 5.4 (CONTINUED) Overview of Semiautonomous Smart Wheelchairs

System

Functions

Control Method

Feedback

Sensors

Navigation

Collaborative Wheelchair Assistant (CWA) (Zeng, Burdet, and Teo 2008; Zeng, Teo et al. 2008)

Navigation

assistance

User controls speed; wheelchair mostly controls heading to reach user-defined destination

Two options for user to deviate from path (collision avoidance): elastic path controller and graphical user interface

None

Encoders Scanner (bar code on floor)

Virtual path learned with walk-through programming (WTP)

Path acts as tracks Elastic path controller: user can deviate but feels attraction to path

IATSL Intelligent Wheelchair System (IWS) (How 2010)

Obstacle

avoidance

Navigation

assistance

Stops if user too close to obstacle

Audio (provides navigation prompts after chair has stopped)

Stereovision

camera

Occupancy grid with blob detection for obstacle avoidance

Direction of greatest freedom computed from occupancy grid for prompting

(Continued)

System

Functions

Control Method

Feedback

Sensors

Navigation

JiaoLong wheelchair (Chen and Agrawal 2013)

Collision

avoidance

Three modes:

  • 1. Manual
  • 2. Minimum vector field histogram (MVFH) aided (obstacle avoidance)
  • 3. Dynamic shared control (DSC) algorithm: weighted control between user and wheelchair

None

Laser range finder

Camera

Encoders

Evaluation of performance in real time to determine user control weight with minimax multiobjective optimization algorithm based on

  • 1. Safety
  • 2. Comfort
  • 3. Obedience

Levels-of-automation (LOA) wheelchair (Jipp 2013)

Collision

avoidance

navigation

assistance

Three modes (LOAs):

  • 1. Manual
  • 2. Collision avoidance with eye tracking to determine intended direction; manual navigation and path planning
  • 3. Autonomous driving based on user gaze to determine target; user-confirmed destination

Touch screen (mode 3)

Eye tracker Ultrasonic sensors

Autonomous navigation with A algorithm: shortest path between position and goal Added button: pressed continually to move (mode 2) and emergency stop (mode 3)

TABLE 5.4 (CONTINUED) Overview of Semiautonomous Smart Wheelchairs

System

Functions

Control Method

Feedback

Sensors

Navigation

NavChair (Levine et al. 1999) behavior replicated in Hephaestus Smart Wheelchair System (Simpson, Poirot, and Baxter 2002)

Obstacle

avoidance

Navigation

assistance

Three modes:

  • 1. General obstacle avoidance (GOA)
  • 2. Door passage (DP)
  • 3. Automatic wall following (AWF)

None

Array of 12 ultrasonic sensors Encoders

Certainty grid to map obstacle locations and find direction with minimal obstacles closest to user input using MVFH Combined with vector force field (VFF) to avoid obstacles in chosen direction

Navigation and Obstacle Avoidance Help (NOAH) (Viswanathan 2012)

Obstacle

avoidance

Navigation

assistance

Stops if user too close to obstacle

Audio prompts

Stereovision

camera

Occupancy grid (obstacle detection) from camera Directional Control Logic Module (DCLM) to interrupt joystick message (stop)

Internal map with Simultaneous Localization And Mapping (SLAM)

Partially Observable Markov Decision Process (POMDP) to estimate user’s cognitive state and issue adaptive prompts

TABLE 5.4 (CONTINUED) Overview of Semiautonomous Smart Wheelchairs

System

Functions

Control Method Fee

dback Sensors

Navigation

Smart Chair (Parikh et al. 2007)

Obstacle

avoidance

Navigation

assistance

Three modes: None

1. Shared (chair drives until user takes control of the

j oystick—checks consistency with goal)

  • 2. Autonomous
  • 3. Manual

Cameras: omnidirectional, human interaction Laser range finder IR sensors (back) Encoders

Three types of actions:

  • 1. Deliberative path planning
  • 2. Reactive obstacle avoidance
  • 3. User initiated Visual interface to choose

destination (video projection) Dead reckoning (encoders) + visual feedback (camera and markers on ceiling); triangulation with recognized landmarks

Adaptable to user: level of autonomy, sensitivity

but industry reports predicted that the additional price a consumer will pay for self-driving car technology will be approximately $5,000 by 2030 (IHS Automotive 2014; Mosquet et al. 2015). Because approximately two- thirds of the cost of the technology underlying self-driving cars is actually devoted to geographic positioning system (GPS) technology (Davies 2015), the additional price of technology for a smart wheelchair might be as low as $2,000.

 
Source
< Prev   CONTENTS   Source   Next >