The dynamic behavior of an inverted pendulum platform inherently reveals information about its interaction with the environment. The closed loop controller puts the unstable plant into dynamic equilibrium, where very small external forces result in persistent excitation of the system. Our previous work has shown that we are able to detect and estimate external forces on an inverted pendulum base using only dynamic state information from the controller. The Yale Inverted Pendulum Door Opening Robot (YIPDOR) has gained a simple 4 degree of freedom (DOF) arm with the goal of opening and traversing doorways with both heavy and sprung doors. We have explicitly omitted dedicated force sensors. Estimation methods using only the Jacobian and joint torques do not provide adequate estimates of external forces on this system. While improved force estimation continues to be a goal for the project, we propose that traversal of sprung doors may be accomplished with low-level, dynamic postures and detection of changes in dynamics rather than explicit force estimation and control. Changes in the equations of motion and dynamic constraints (contact, unlatching, door movement) are detected by using filtered derivatives of relevant parameters. This feature extraction research increases the safety and performance of the balancing and manipulation system without any additional dedicated sensors.

In this paper a method for force estimation on inverted pendulum type balancing platforms is presented and examined. Inverted pendulum systems are sensitive to small external forces because they are being balanced about an unstable position of zero force. The perturbations that result from external forces cause noticeable changes to the system state variables and it is these changes that are used to estimate the force. We present a dynamic model for the inverted pendulum and then examine several methods for estimating an external force. Use of the open loop plant parameters offers the best method for estimating the net torques acting on the system. A significant outcome is that estimation can be performed using only the system mass properties along with the inertial and displacement sensors that are required for balancing and this estimate is more accurate than a comparable drive system in a statically stable robot. Thus improved one axis force estimation may be accomplished using the same sensors that are used to stabilize the platform.