Kognitive Systeme

MobilAssist: Intelligent Wheeled Walker

Independent and self-determined living is a valuable good that should be sustained as long as possible. For the elderly or people with a visual or cognitive impairment this is not taken for granted as easily. Rollators are a commonly used motion aid in the everyday life of elderly people. Making these aids more intelligent in order to not only support their physical abilities, but also to support their visual and cognitive capabilies, could enable them to maintain an active life and reduce the amount of help required from caregivers.

The beActive+e wheeled walker used for this project is developed by our project partner Bemotec GmbH. It has two powered back wheels which support the user in their motion. With a maximum range of 30km at a maximum speed of 0.8m/s it is capable of supporting a user over longer distances and handling a variety of everyday tasks. The intelligent prototype is equipped with an onboard computer, a RGB-D camera, a stereo camera, a 2D laser scanner, a GPS module and a touchscreen.

Fig. 1: The intelligent beActive+e prototype

Project Goals

Obstacle Detection

Since safety is regarded the most important aspect, situations which may pose a threat to the user have to be recognized and avoided in advance. The prototype is able to detect and avoid negative and positive obstacles. Negative obstacles are obstacles below the ground plane, e.g. sidewalk edges or downward leading stairways. Positive obstacles are above the ground, e.g. furniture or persons. Using its RGB-D or stereo camera a local map of the close by environment is created and used to identify both negative and positve obstacles very precisly.

Traversability Analysis

A detailled traversability analysis can remotely predict several safety relevant criteria for the vehicle and it's user. These criteria include: The 3D orientation of the rollator, the wheel support and the angular velocity. Based on these criteria different actions can be performed: Warning the user, stopping the vehicle or planning a safe path as close to the user's intended motion as possible.

Fig. 2: The vehicle model in a test environment with several obstacles. Green lines represent feasible paths. The brown line indicates the target direction. The purple line is the path selected for driving.

Video

https://youtu.be/mihJ732VpY4

Shared Control

Due to the direct physical interaction of the user with the wheeled walker, human-robot-interaction is a highly relevant topic for this project. Restrictions to the user's intended motion should be as small as possible and interventions by the algorithm should have reasonable causes to achieve a high user acceptance.

Navigation

The navigation system provides guidance to the user for in- and outdoor areas. For outdoor navigation GPS is used in conjuction with external map data for path planning and localization. In indoor areas a SLAM system is used to build a map and perform localization based on the 2D laser scan data.

Fig. 3: Path planning on OpenStreetMap data.
Fig. 4: Indoor localization and mapping with the 2D laser scanner.

Project Partner:

Bemotec GmbH, Reutlingen

This work is funded by the Federal Ministry of Education and Research under research grant number 01IS15049B.

References

[1] Julian Jordan and Andreas Zell. Real-time pose estimation on elevation maps for wheeled vehicles. In Intelligent Robots and Systems (IROS), 2017 IEEE/RSJ International Conference on, Vancouver, Canada, September 2017. (accepted for publication). [ details | pdf ]
[2] Julian Jordan and Andreas Zell. Kinematic model based visual odometry for differential drive vehicles. In IEEE 8th European Conference on Mobile Robots (ECMR), ENSTA ParisTech, Université Paris-Saclay, France, 2017. (accepted for publication). [ details | pdf ]
[3] Julian Jordan and Andreas Zell. Ground plane based visual odometry for rgbd-cameras using orthogonal projection. IFAC-PapersOnLine, 49(15):108 -- 113, June 2016. 9th {IFAC} Symposium on Intelligent Autonomous Vehicles {IAV} 2016, Leipzig, Germany, 29 June-1 July 2016. [ DOI | details | link ]