My Current Research
The
Concept and Algorithm
At Georgia Tech my current research involves the study of remote teleoperation via the Internet. In a basic bilateral teleoperation setup with force feedback, the user controls a master joystick that sends position signals to slave robot. The slave robot attempts to accomplish the task, and as it interacts with the environment it sends feedback force signal towards the master robot. Under normal circumstances, the only component that is generating energy is the human user at the master end. However, if there is a transmission delay, the transmission line may become an energy-generating component and hence violates passivity. My research involves looking into various control strategies and algorithms that can counteract this problem. This research is applicable in any remote-control devices where transmission lag is significant, including remote-control robots over space, ocean depths via sonar, and over the internet.
Conventional Bilateral Teleoperation
The current control scheme used to solve the problem involves using wave variable before and after transmission. Wave variables in a sense turns velocity and force signals into "wave" signals that have the dimensions of square-root of power and hence can be considered as representative of energy. If the wave signal is delayed across the transmission line, energy is simply stored. This ensures passivity of the overall teleoperation system. However, simulations and experimental results have shown poor transient response. Hence a predictor is used to estimate the output of the slave system based on a predetermined slave model. The states of the slave are sent to the predictor on the master side via the transmission line. However, these states are outdated due to the transmission delay. The predictor then marches these states from the past to the present to estimate the present slave output, effectively faking the master into thinking there is no delay. This dramatically improves the transient response of the system. But due to variations in transmission delay as well as predictor model inaccuracy, the master and slave positions may drift apart. Hence a drift correction algorithm is added on the slave side that attempts to drive the slave desired to master position via the wave variable.
Wave
Variables Only Wave Variables with Predictor
and Drift Control
Furthermore, the slave environment may change over time requiring an adaptive predictor. A common situation is when the slave moving from free space into rigid contact. Two methods are explored to deal with such situation: a semi-adaptive predictor that attempts to identify the location of rigid contact switching between free space and rigid contact modes accordingly, and a full-adaptive predictor that attempts to identify and adapt to the environmental force based on a recursive least square (RLS) algorithm. The semi-adaptive predictor is superior in free space-rigid contact transition because once the predictor identified the point of contact it can anticipate when the slave will leave contact. The full-adaptive is more robust because it is not limited by a locked position as it only tries to identify the environmental force and parameters, hence it can adapt to other scenarios such as loaded slave.
Semi-Adaptive
Predictor under Rigid Contact Full-Adaptive
Predictor under Rigid Contact
Videos demonstrating the algorithms in action are shown here: unstable, wave variables only, wave variables with predictor,
wave variables with predictor and drift control.
The Hardwares
We have purchased two Phantoms by Sensable to test the wave-based control algorithms at the hardware level. The Phantom is a haptic device with three degrees of freedom. The results shown above are all produced via these devices. The two Phantoms are run by computers using Windows 2000 operating system at approximately 1 kHz. The two machines are connected via the internet using user datagram protocol (UDP). Transmission delay is simulated with a lower limit of 400 ms and an upper limit of 470 ms with a 10% chance of data loss.
With the success of testing the algorithms on the Phantom, we have proceeded to expand the research to incorporate a hydraulic actuated lift (HAL) as the slave device. HAL, in contrast with the electromechanical Phantom, is a hydraulic device with flow control servo-valves with two degrees of freedom and a workspace of roughly five times that of the Phantom. By attempting to control HAL with the Phantom with transmission delay, we can evaluate the effectiveness of the algorithms in a highly asymmetric (dynamically and in scale) master-slave setup. Wave variable algorithm in this setup has already been shown to work well. Initial testing of wave variable with predictor and drift control has yielded promising results.
Hydraulic Actuated Lift (HAL) Phantom