1. Exploiting Natural Dynamics in the Control of a 3D Bipedal Walking Simulation Jerry Pratt, Gill Pratt MIT Leg Laboratory www.ai.mit.edu/projects/leglab/

2. M2 3D Biped -- 12 DOF Designed by Daniel Paluska and David Robinson

3. M2 Simulation 3D Biped -- 12 DOF

4. Spring Flamingo Planar Biped 6 DOF

5. Outline Background:Background: –Previous Work –Series Elastic Actuators Natural Mechanisms for WalkingNatural Mechanisms for Walking Planar Robot Algorithm and ResultsPlanar Robot Algorithm and Results 3D Simulation Algorithm and Results3D Simulation Algorithm and Results Conclusions and Future WorkConclusions and Future Work

6. Background: Previous Work

8. Passive Dynamic WalkersPassive Dynamic Walkers –Natural Looking –Efficient –Simple (No) Control Powered Bipedal Robots –Capability Potential –Practical/Market Potential –Biological Similarity Potential Passive Walkers Powered Walkers

9. Background: Series Elastic Actuators Motor Gear Train Series Elasticity Load Force Source Load

10. Background: Series Elastic Actuators

11. Designed by David Robinson and Daniel Paluska

12. Outline BackgroundBackground Natural Mechanisms for WalkingNatural Mechanisms for Walking –Kneecap –Compliant Ankle –Passive Swing Planar Robot Algorithm and ResultsPlanar Robot Algorithm and Results 3D Simulation Algorithm and Results3D Simulation Algorithm and Results Conclusions and Future WorkConclusions and Future Work

13. Kneecap

14. Kneecap

15. Compliant Ankle

17. Passive Swing

18. Outline BackgroundBackground Natural Mechanisms for WalkingNatural Mechanisms for Walking Planar Robot Algorithm and ResultsPlanar Robot Algorithm and Results 3D Simulation Algorithm and Results3D Simulation Algorithm and Results Conclusions and Future WorkConclusions and Future Work

19. Spring Flamingo Planar Biped 6 DOF

20. Planar Robot Algorithm

21. Height: Constant vertical force.Height: Constant vertical force. Pitch: PD controller on stance leg.Pitch: PD controller on stance leg. Swing Leg: Mostly Passive.Swing Leg: Mostly Passive. Speed: Naturally stable for slow speeds.Speed: Naturally stable for slow speeds.

22. Robot Strobe Photo

23. Planar Robot Animation

24. Physical Robot Walking Data

25. Outline BackgroundBackground Natural Mechanisms for WalkingNatural Mechanisms for Walking Planar Robot Algorithm and ResultsPlanar Robot Algorithm and Results 3D Simulation Algorithm and Results3D Simulation Algorithm and Results Conclusions and Future WorkConclusions and Future Work

26. M2 Simulation 3D Biped -- 12 DOF

27. Simple Foot Placement Model

29. “Capture Angle”

30. 3D Simulation Algorithm

31. 3D Simulation Animation

32. Simulation Data

33. Outline BackgroundBackground Natural Mechanisms for WalkingNatural Mechanisms for Walking Planar Robot Algorithm and ResultsPlanar Robot Algorithm and Results 3D Simulation Algorithm and Results3D Simulation Algorithm and Results Conclusions and Future WorkConclusions and Future Work

34. Conclusions Passive Mechanisms Helped Make ControlPassive Mechanisms Helped Make Control –Simple –Efficient –Natural Looking Actuators with Negligible Dynamics are ImportantActuators with Negligible Dynamics are Important Decoupled Controller Works well for 3D Simulation Walking StraightDecoupled Controller Works well for 3D Simulation Walking Straight

35. Future Work Faster WalkingFaster Walking More Robust AlgorithmMore Robust Algorithm Biologically Similar AlgorithmBiologically Similar Algorithm Automatic Tuning, LearningAutomatic Tuning, Learning Central Pattern GeneratorsCentral Pattern Generators 3D Robot3D Robot

36. ? What Next?