Jesse Pentzer

Model-based Prediction of Skid-steer Robot Kinematics Using Online Estimation of Track Instantaneous Centers of Rotation

This paper presents a kinematic extended Kalman filter EKF designed to estimate the location of track instantaneous centers of rotation ICRs and aid in model-based motion prediction of skid-steer robots. Utilizing an ICR-based kinematic model has resulted in impressive odometry estimates for skid-steer movement in previous works, but estimation of ICR locations was performed offline on recorded data. The EKF presented here utilizes a kinematic model of skid-steer motion based on ICR locations. The ICR locations are learned by the filter through the inclusion of position and heading measurements. A background on ICR kinematics is presented, followed by the development of the ICR EKF. Simulation results are presented to aid in the analysis of noise and bias susceptibility. The experimental platforms and sensors are described, followed by the results of filter implementation. Extensive field testing was conducted on two skid-steer robots, one with tracks and another with wheels. ICR odometry using learned ICR locations predicts robot position with a mean error of -0.42i¾?m over 40.5i¾?m of travel during one tracked vehicle test. A test consisting of driving both vehicles approximately 1,000i¾?m shows clustering of ICR estimates for the duration of the run, suggesting that ICR locations do not vary significantly when a vehicle is operated with low dynamics.

On-line estimation of vehicle motion and power model parameters for skid-steer robot energy use prediction

This paper presents a method of estimating skid-steer robot power usage using on-line estimation of terrain and kinematic parameters. For vehicles operating at low speeds on hard, flat surfaces, kinematic models utilizing the instantaneous centers of rotation (ICRs) of the tracks or wheels of a skid-steer vehicle have been shown to provide accurate motion and power use estimation. Previous work has relied on post-process optimization to learn necessary ICR location and terrain information for motion and power modeling. The work presented here utilizes an extended Kalman filter for learning ICR locations and the recursive least squares algorithm for learning terrain-related power model parameters. The algorithms have been implemented on a wheeled skid-steer vehicle, and field test results show good estimation of motion and power usage using no prior terrain information and only knowledge of vehicle geometry and mass distribution, intermittent GPS and heading, and odometry information from the slipping tires/treads.

The use of unicycle robot control strategies for skid-steer robots through the ICR kinematic mapping

While decades of work and hundreds of research papers exist on unicycle robot control, the control of skid-steer robots is not yet as standardized due to the complexity of wheel slipping behavior. This work presents a method of utilizing the track or wheel Instantaneous Centers of Rotation (ICRs) on a skid-steer vehicle to map skid-steer dynamics to an equivalent time-varying model of unicycle dynamics. This allows for the direct implementation of existing unicycle, or Hilare type, robot trajectory controllers. Knowledge of ICR locations enables the calculation of required track or wheel speeds to create desired vehicle movement, similar to the kinematic relations resulting from the no-slip assumption of a unicycle robot. The algorithm requires no prior knowledge of vehicle dimensions or terrain parameters because ICR locations are estimated during robot operation using an extended Kalman filter (EKF). Simulation and experimental results for a wheeled skid-steer vehicle show good trajectory tracking performance.

Extended kalman filter for improved navigation with fault awareness

Most unmanned mobile robotic platforms contain multiple sensors that can be leveraged to measure vehicle motion states, where there often exists redundancies among the different sensor types. Kalman filter based sensor fusion between inertial navigation sensors, GPS readings, encoders, etc. is a very popular approach in the literature to improve the accuracy of navigation readings. However, such redundancies can also be exploited for simultaneously conducting fault detection and identification of the sensors and the robot. This paper presents theory and results for an Extended Kalman Filter (EKF) approach fusing IMU/INS readings with GPS and/or visual odometry (VO) data to diagnose faults in wheel odometry readings (encoders). A key advantage is that the approach works for detecting faults, even when relatively low grade and inexpensive sensors are installed in the vehicle.

Energy-based path planning for skid-steer vehicles operating in areas with mixed surface types

This paper presents a method of path planning for skid-steer robots using an energy-based heuristic. A kinematic model of skid-steer motion utilizing the instantaneous centers of rotation (ICRs) between the tracks and the ground surface is used to predict vehicle motion. A model of skid-steer robot power usage, which also utilizes ICR estimates for slip velocity calculation, is implemented to generate estimates of energy usage. The kinematic and power use models are fused with a Sampling Based Model Predictive Optimization algorithm to plan energy efficient paths through operational areas with mixed surface types. The results of planning paths through both simulated and real-world environments are presented and show that small increases in distance can result in significant energy savings for skid-steer robots.

SOM with neighborhood step decay for motor current based diagnostics

Embedded self-learning is a desired capability that can enhance autonomy in different types of unmanned systems. Autonomous diagnostics is an area of opportunity to deploy this capability, which allows for vehicle failure awareness and enables for other advantageous schemes such as fault tolerant control. In this paper, we present one subsystem of an ensemble of schemes that form the Enhanced Autonomous Health Monitoring System (EAHMS) designed to support NASAs Robotics, Tele-Robotics and Autonomous Systems Roadmap. The EAHMS is aimed to provide an integral framework to determine the operational condition of on-board sensors (odometry), actuators, and power systems. Within the EAHMS context, this paper outlines a method for diagnostics of a robotic vehicle mechanical mobility subsystem by motor current and vibration signature analysis based upon Self Organizing Maps (SOM) using an enhanced neighborhood step decay algorithm. The learning algorithm was tested for different learning rate functions and was applied to different training set cases. The resulting algorithm was used for conducting failure diagnostics in a testbed, where three types of transmission/motor mechanical failures were considered: (a) damaged chain link; (b) motor gearbox damage; and (c) damaged sprocket. A core goal of this diagnostic approach is to enhance a novel methodology called the embedded Collaborative Learning Engine (eCLE), which combines supervised and unsupervised learning synergistically to process new emerging data signatures. This technique for system enhancement and application results are described in this paper.

Improving autonomous underwater vehicle navigation using inter-vehicle ranging

One-way-travel-time (OWTT) acoustic ranging has received considerable attention as improvements to acoustic modems and electronic clocks have made it a feasible navigation tool. This paper reports the results of simulations investigating the effect of utilizing inter-vehicle ranging for autonomous underwater vehicle (AUV) navigation. In these simulations, a fleet of AUVs operates in shallow water with a pair of fixed transponders. A rigid timing cycle for acoustic communications was implemented with a message queuing approach to simulate the handicaps of underwater acoustic communication. Furthermore, a simple path following algorithm was used to navigate the AUVs through a waypoint course, and a kinematic motion model was used to simulate AUV movement. The position of each vehicle in the fleet was estimated independently by combining the propagation steps of an extended Kalman filter with the update equations of an extended information filter. Results of the simulations showed the addition of inter-vehicle ranging improved accuracy by 1-2 cm when navigating using four fixed transponders and by 9-24 cm when navigating using two fixed transponders.

Investigation of the effect of continuously variable transmissions on ground robot powertrain efficiency

Explosive ordinance disposal (EOD) robots are limited in endurance and range by the amount of energy available in the batteries used to power them. Continuously variable transmission (CVT) technology has developed quickly in recent years in the automotive field and is now being applied to smaller vehicles such as bicycles and electric scooters. This paper will discuss simulations investigating the feasibility of adding CVT transmissions to robot powertrains in order to improve the overall efficiency of the drive system. The equations used to calculate the power required to move a robot at varying speeds will be described, as well as the equations used to model CVT and direct-drive transmissions, DC motors, and power discharge from a battery. The results of a constant mass simulation, where the added CVT mass was offset by a loss in battery mass, showed that adding CVTs is not a feasible option due to the mass of the CVTs. In an added mass scenario, where the mass of the CVTs was added to the overall robot mass, the benefits of a CVT depended strongly on the speeds at which the robot was expected to perform. A robot expected to operate at low speeds most of the time would benefit more from a CVT than a robot expected to operate near maximum speed most of the time.