Internal posture sensing for a flexible frame modular mobile robot

Abstract

-A sensor fusion algorithm for flexible framed modular mobile robots is presented in this paper. This algorithm uses traditional Kalman filters and rigid axle kinematic models to predict the global posture of each axle. A Covariance Intersection filter is then proposed for fusing these axle modules using data provided by the compliant frame module.Modeling and instrumentation of the compliant frame module is the remaining focus. The instrumented frame is required to estimate the relative posture of the axles as well as the force components and moment that the beam exerts on each axle. These estimates must also be valid for large deflections in order to accommodate a reasonable turning radius. To accomplish these goals the beam equations are derived. A linear interpolation of the strain gauge data is used to calculate posture, force, and moment estimates. Experimental results show that the frame module can yield accurate relative posture estimates for large deflection.

Instrumentation and Algorithms for Posture Estimation in Compliant Framed Modular Mobile Robots

“…P R P S is used during the second tier data fusion described in Section 5 and is evaluated experimentally in Section 6. The RPS algorithm can also predict forces applied to the frame, but since this information is not used here, the reader is referred to Merrell (2006) or Merrell and Minor (2003) for more information.…”

“…Traditional sensors, such as odometry, inertial measurement, or GPS are insufficient for this task since drift cannot be reduced sufficiently. Thus, in this research we implement a Relative Posture Sensor (RPS), introduced by Merrell and Minor (Merrell and Minor 2003), to produce more accurate relative posture estimates and reduce drift between the axles.…”

“…In the RPS, a polynomial of the same order as number of strain gages is used to directly fit strain gage data. Assuming a linear stress-strain relationship, we then derive a curvature equation that is piecewise integrated to estimate deflection and force (Merrell and Minor 2003). Research by Piedboeuf and Miller (1995) similarly applies a polynomial to directly determine tip displacement and orientation.…”

“…Since the deflections of our system are highly nonlinear (as much as 275 3 in extremes) and the system may be in post-buckling, such modal based approximations are not useful. While we build upon the work by Piedboeuf and Miller (1995) to also obtain force estimates in Merrell and Minor (2003), we only use position and orientation data in this research since our focus is on localization. Interaction forces between the axles is considered in other research for axle control (Park and Minor 20041 Zhu, Minor and Park 2006), but we apply a post-buckled stiffness model for simplicity.…”