@import url(http://kodlab.seas.upenn.edu/pub/skins/sinorca/basic.css); @import url(http://kodlab.seas.upenn.edu/pub/skins/sinorca/layout.css); @import url(http://kodlab.seas.upenn.edu/pub/skins/sinorca/sinorca.css);
Body state estimation project aims to develop a heterogenous sensory system for a RHex class robotic platform operating in a dynamic regime. Our virtual sensor estimates the full body states, 6-DOF configuration and their first time derivatives, in real-time. Our implementation employs two major sensory systems: 1) tactile sensors embedded in the compliant legs; and 2) inertial sensors embedded in the body. The sensory data stream from these sources are fused to produce body state estimation with significantly higher accuracy than any one of the sensors can provide alone.
The tactile sensors embedded in the compliant legs measure the kinematic configuration of the passively actuated degrees of freedom of legs. When certain conditions are met this information can lead to a very accurate estimate of body pose. The applicability of this approach to body state estimation requires: 1) known ground geometry; 2) no foot slippage; and 3) at least three non co-linear support legs. The free rotating legs in RHex does not permit an effective way to transfer sensitive analog measurements over physical channels. To remedy this issue we designed LegNet?, a wireless short range sensory network.
The inertial sensor embedded in the body provides acceleration and rotational speed measurement in the body coordinate system. In some of our implementations we considered physically distributed sensory constellations. Employing a dynamic rigid body model such inertial measurements can provide us with a finite time horizon estimate of body pose. However, inertial measurement based pose estimation is bound to diverge to noise integration. Especially in a hybrid system such as RHex where the ground contact events produce impulsive ground reaction forces the efficacy of the inertial method becomes severely reduced. On the other hand, inertial approach is very well suited during aerial phases in a dynamic operation.
This work was done during my post at the University of Michigan.
Copyright Kodlab, 2017