An active orthosis for cerebral palsy children - Emanuele Bonanni e Andrea Calanca

Transcript

1. 11/05/12 A.Calanca 1 A.Calanca Sept 24, 2012 An Active Orthosis For Cerebral Palsy Children A. Calanca, S. Piazza, P. Fiorini, A.Cosentino ALTAIR Robotics Laboratory Computer Science Department University of Verona
2. Background: Cerebral Palsy Cerebral palsy has an incidence of birth between 0.15 and 0.25%. Precocity of rehabilitation has a fundamental role in prevention of secondary deformity and anomalous motor development [Viurtello1984]. Also recent studies pay great attention on physical therapy applied to young CP patients, focusing on movement based strategy and physical training [Dodd 2002][Damiano 2006]. This kind of treatments are quite expensive because they need the presence of one or more physiotherapists. Orthotic systems try to help this treatment relieving physiotherapist of part of work. A.Calanca Sept 24, 2012
3. The ARGO Prototype Sensors: • Muscle force • Hip and knee angles • Ground reaction forces Actuation: • Pneumatic Muscle • Force Control • Reciprocation A.Calanca Sept 24, 2012
4. The ARGO Prototype A.Calanca Sept 24, 2012
5. The ARGO Prototype Interaction with the test patient A.Calanca Sept 24, 2012
6. Results Patient condition before active orthosis A.Calanca Sept 24, 2012
7. Results Autonomous walking: the user is able to keep a fluid walking and also start it without external help. A.Calanca Sept 24, 2012
8. Results Rehabilitation: our test patients show a gradual improvement of his motion capability. A.Calanca Sept 24, 2012
9. Actuation System In particular we use Festo manufactured muscles: they have an unique layer of mixed rubber and fibers that improves muscle safety and long lasting in respect with classical McKibben. Disadvantages: non-linear behaviour, control issue. McKibben Pneumatic Artificial Muscles • Intrinsic safety and compliance • High power to weight ratio • High forces • Low cost • Supply via small high pressure air bottle. A.Calanca Sept 24, 2012
10. Chou and Hannaford model for classic McKibben muscles: F is the force, P is the pressure and θ is the fiber angle. We can put the same model in a different form involving muscle length (L) instead of θ, which is difficult to measure This is more convenient for identification! Actuator modelling A.Calanca Sept 24, 2012
11. Actuator modelling Chou and Hannaford model is not suitable for Festo muscles, due to different mechanical structure. Validation result (LS identification - linear parameterisation): A.Calanca Sept 24, 2012
12. Neural Network Type: feed-forward, back propagation Topology: 8 neurones (2 input, 5 middle, 1 output) Training Data collected from test bed experiments at different pressure levels and different force frequencies. Pressure and muscle length as input, force as target Control System A.Calanca Sept 24, 2012
13. We use a neural network (NN) feed-forward action for ensure controller fast response and a low gain PID for stabilisation. The NN calculates the required pressure knowing the Force reference and the muscle length Control System Note: The muscle model M is coupled with the mechanical systems dynamics (DYN) through muscle length A.Calanca Sept 24, 2012
14. Control System Square and sine wave response of the proposed controller. Maximum overshoot of step response is 0.87N while maximum sine following error is 1.31N. Average errors are 0.15 (square) and 0.37N (sine). A.Calanca Sept 24, 2012
15. Control System Comparison with low gain PID: the response is stable and not noisy and but following errors are very big A.Calanca Sept 24, 2012
16. Control System Comparison with high gain PID: the response is fast but is too noisy. Note: There is no usable compromise between the showed low gain and high gain configurations! A.Calanca Sept 24, 2012
17. Control System Response of the proposed controller in orthosis usage. The maximum error is of 1.31N. Note: The human leg has more filtering action with respect to the test bed. Some oscillation still occurs but they are independent from set point dynamics as we expect A.Calanca Sept 24, 2012
18. It generates the hip torque profiles basing on sensor input and system knowledge. It uses a simple algorithm for gait phase recognition, based on a finite state machine (FSM). Torque Computation A.Calanca Sept 24, 2012
19. Torque Computation Then accordingly to FSM state, we calculate the desired torque basing on gravity compensation and the equilibrium of the patient. A.Calanca Sept 24, 2012
20. Results Torque, position and FSM state data from session with test patient. The system is not cohercitive and is able to understand user intention. Plot shows FSM states in a double left step. A.Calanca Sept 24, 2012
21. Results Patient condition before active orthosis Two years of treatment with passive orthosis (same mechanical structure). Patient is not autonomous in walking and needs help from physiotherapist. A.Calanca Sept 24, 2012
22. Conclusions Experiments with a cerebral palsy patient show very encouraging results. He was not only able to walk autonomously but also to improve his capability in passive orthosis usage. This can be due to the interaction with the orthosis that doesn’t make the user passive but follows his action plan. Patient action plan Vs Physiotherapist action plan A.Calanca Sept 24, 2012
23. Thanks
24. Results The active orthosis does not produce a significative decrease in user fatigue and in muscle recruitment • Metabolic cost analysis • EMG analysis
25. Future Works We need to carry on experiments with more CP patient for having more scientific evidence of benefits We propose to investigate if there is an improvement in patient condition. How much is the improvement? In wich aspects? Is it dependent from patient initial condition? How? What can be the best way to help patient? Are all still open question.
26. Result Autonom ous Start (3 trials) Continuo us Walking Duration NF- Walker (before ARGO) 0/3 3-5 s (with little help) ARGO 3/3 >200 s NF- Walker (after 3/3 >200 s
27. Background: Active orthoses Table shows that there is not significant active orthosis which is mobile and can keep user balance. Note: Only most famous rehabilitative/assistive devices are included in table. Self Balance d Not Self Balance d Mobile RewalkT M, AAFO, WWH, HAL Not mobile Lokomat ®, Gangtrai nerTM,