Research On Joint Angle Measurement For Soft Exosuit Based On Multimodal Fusion

Exoskeleton robot has been successfully applied in more and more scenes as a kind of wearable intelligent equipment.Rapid development can be seen in these years.The main body of exosuit is made of soft textiles,which is very light and flexible.This kind of equipment is able to offer remarkable assistance force to the wearer to save energy and lower metabolic cost rate.However,due to the flexibleness,soft exosuit is extremely hard to model and thus very hard to estimate its joint angle using traditional ways.Troubles such as random binding of straps,motion impact,and changing application scenes make joint angle measurement even harder.So we need to develop a more suitable detection method to obtain real-time joint angle of soft exosuit.This work aims on developing a soft exosuit joint angle measurement system including hardware and algorithm to estimate pose accurately and conveniently.The angle measurement system is developed based on SIAT Soft Exosuit-III platform,using multimodal information from accelerometer,plantar pressure sensor,gyroscope,and machine learning technology.The system solves key problems such as calibration procedure which is hard to simplify,binding error which is hard to eliminate,motion impact which disturb accelerometer's measurement,accumulated error of gyroscope and so on.Finally,by combining the advantages of multimodal information fusion technology and optimization based on the application scenes of soft exosuit,a complete joint angle measurement system is proposed for soft exosuit.This work mainly includes the following contents:1.The introduction of soft exosuit used in this project,including mechanical principle and related control system.The introduction of hardware system for joint angle measurement,including sensors and hardware circuit.And the introduction of optical motion capture system which is taken as ground truth of this work.2.The designing of GTPN neural network architecture.By combining the basic idea of auto-encoder and generative adversarial network,we can predict the wearer's hip joint angle trajectory.This technology can help us obtain a smooth prediction of joint angle which is as close as possible to the ground truth.3.Eliminating the binding error of soft exosuit by multimodal information fusion.This method can improve the adaptability of soft exosuit to different binding ways and different body shapes by simply doing some free movements.4.Using multimodal fusion to estimate joint angle by accelerometer measurement,and fuse it with GTPN prediction result to improve accuracy and robustness to motion impacts.Using region pressure differentiate method to detect key poses and divide gait cycles based on plantar pressure sensors.Integrating angle velocity to obtain joint angle in a gait cycle.Finally,complementary fusion is used to fuse all the information to obtain a better joint angle measurement.