Gait Controller for 3D Active Dynamic Walking

ECE 780 Graduate Course Project Presentation

Transcript

1. Gait Controller for 3D Active Dynamic Walking Safwan Choudhury ECE 780: Project Presentation
2. Motivation ▪ Bipedal Walking ▪ Human-like Gait ▪ Limited power source ▪ Energy efficiency 2
3. Motivation ▪ Passive Dynamic Walkers ▪ Energetically efficient ▪ Human-like Gait ▪ Limited Region of Stability ▪ Dependence on sloped incline 3
4. Motivation ▪ Virtual Passive Dynamics ▪ Concept of virtual gravity ▪ Minimal actuation ▪ Functionality on level ground 4
5. Motivation ▪ Virtual Passive Dynamics ▪ Virtual gravity torque ▪ Applied to joints in the sagittal plane ▪ Angle Φ is represents the angle of inclination of the slope 5 τ virtual = m H l + ma + ml ( )cosθ 1 −mbcosθ 2 ⎡ ⎣ ⎢ ⎢ ⎤ ⎦ ⎥ ⎥ gtanφ
6. Related Work ▪ Passivity based control (Spong) ▪ Energetics of active powered locomotion (Kuo) ▪ Gait generation and control techniques (Asano) ▪ Energy constraint control (Asano) ▪ Use of semi-circular feet (Asano) 6
7. Project Goal ▪ Extending active dynamics beyond 2D ▪ Decoupled sagittal and frontal dynamics ▪ Simulation and comparison with strictly sagittal case ▪ Simulation and comparison with conventional 7
8. Kinematics ▪ Compass Gait Kinematics ▪ Total link length: l = a + b ▪ Total Mass: M = mh + 2m ▪ Ankle 1 fixed to (0,0) 8
9. Frontal Dynamics ▪ Equation of motion derived from Lagrangian ▪ Ankle roll torques applied to keep compass biped from falling over 9 M(θ) θ + C(θ,  θ) θ + g(θ) = τ roll + JT F ground A(θ) = M(θ) b(θ,  θ,τ,F) = C(θ,  θ) θ + g(θ) − τ roll − JT F ground A(θ) θ = b(θ,  θ,τ roll ,F ground )  θ roll = A(θ) ( )−1 b(θ,  θ,τ roll ,F ground )
10. Sagittal Dynamics ▪ Common equation of motion ▪ Virtual gravity torque applied to achieve level walking ▪ Similar to frontal dynamics 10 M(θ) θ + C(θ,  θ) θ + g(θ) = τ virtual + JT F ground A(θ) θ = b(θ,  θ,τ virtual ,F ground ) τ virtual = m H l + ma + ml ( )cosθ 1 −mbcosθ 2 ⎡ ⎣ ⎢ ⎢ ⎤ ⎦ ⎥ ⎥ gtanφ
11. Simulation & Control ▪ Started off trying to modify Kajita’s toolbox (RNE) ▪ Running into issues with contact model ▪ Currently writing my own forward dynamics routines ▪ Controller implemented in Matlab Simulink 11
12. 12 To Workspace simout STOP PD Controllers error state torques Gait Trajectory ref Disturbance Detect Fall state Compass Biped torques q qd state Control Strategy
13. 13 Control Strategy torques 1 State Switch state P D PD Logic P D error torques state 2 error 1
14. 14 Control Strategy state 3 qd 2 q 1 Sagittal Forward Dynamics torques forces q qd Ground Contact Model q qd force state Frontal Forward Dynamics torques forces q qd torques 1
15. Expected Results ▪ Lower torque demands in comparison to completely actively-controlled gait ▪ Very limited region of for stable walking, any disturbance will cause it to fall ▪ Sagittal and frontal dynamics are coupled 15
16. Questions? 16