Stair and Ramp Recognition for Powered Lower Limb Exoskeletons

Abstract

Robotic lower limb exoskeletons allow paraplegics to stand up, walk, and overcome obstacles like stairs and ramps. The selection of the required movement pattern is generally done manually by the user, which reduces the ease of use and prohibits dynamic transitions. The proposed solution for this problem is a machine vision based obstacle recognition system, which allows the exoskeleton to adapt its movements to the environment. To achieve this, the system needs to categorize the obstacle, determine its distance, and measure physical characteristics like step height and ramp slope.The developed system uses an infrared stereo camera paired with an IMU to capture a 3D point cloud of the path of the exoskeleton. Using the RANSAC algorithm, planes are segmented from the point cloud and stairs and ramps are identified based on the geometric constellation of the extracted planes. These surfaces are then used to calculate the quantitative characteristics of the obstacle, that would be needed for adaptive trajectory planning of an exoskeleton. To minimize the probability of wrong classifications or measurements, a higher level evaluation was integrated, which rates the quality of the gathered data based on redundant analyses and suppresses wrong results.The system was implemented on low power hardware intended for mobile applications to match the use case of a mobile exoskeleton. In first experiments, the system showed good results in terms of its detection rate and measurement accuracy, but is currently held back by long process run times due to hardware limitations.