Vision-Based Control of an Inverted Pendulum using Cascaded Particle Filters

The inverted pendulum represents an example of a nonlinear and unstable dynamic system whose properties and related control strategies have been studied extensively in the control literature. A basic setup consists of a translation device - the cart - and a rotating boom - the pendulum - which is balanced in its upright position. The control algorithm uses the position of the cart and the rotation angle of the pendulum as input measurements provided by different sensors. While contacting sensor principles such as angle encoders have been used in different demonstrators, we present a purely vision-based tracking system. In combination with state of the art control algorithms our system is able to swing-up and balance the pendulum. The presented setup is non-contacting and does not require specific visual markers on any part of the pendulum. We use off-the-shelf components to realise the monocular tracking system. Experimental results show that our system is able to robustly control the inverted pendulum in real-time on a standard desktop PC.