State estimation for ParkourBot Siyuan Feng and Xinjilefu
Outline • System overview • Problem statement • Proposed solution • Potential problems Potential problems • Questions
System overview • 2 legged dynamic climber, jumps between parallel walls to climb. 5 DOF in total. -Each leg has 2 DOF, retraction and leg angle. -Gyro stabilizer that would passively -Gyro stabilizer that would passively stabilize body orientation during impact, and can change body orientation when actuated. • Periodic hybrid system with alternating stance and flight phase. • Off board sensing and computation. Robot communicates with computer through XBee.
System overview(continued) • An air table is used to reduce friction and impose planar constraints. • Camera mounted on top of air table, captures (x, y) positions. (x, y) positions.
System overview(continued)
Problem statement • Want real time estimation of robot configuration (x, y, θ) and measurement for controls (φ_l, φ_r, r_l, r_r). • Want to predict the state for next impact. • Want to predict the state for next impact. • System identification (damping coefficients for stance phase dynamics, friction). • Environment estimation (wall width).
Proposed solution • EKF for state estimation. • EM for parameter (damping) tuning.
Potential challenges • Crappy camera (frequency, resolution, irregular frame updates, etc) • Unknown parameters in motion model (damping, friction). No analytical solution to (damping, friction). No analytical solution to stance dynamics. • Hard time limits for estimation and prediction.
Recommend
More recommend