Test Beds

Mobile Human Scale Equipment - Compact Rescue Robot (Test Bed 4)

NEW!

See video of the robot test bed undergoing gait testing in June 2010.

Leader

Prof. Wayne Book (Georgia Tech)

Statement of Test Bed Goals

The goal of this test bed is to demonstrate a compact rescue robot, an example of portable, un-tethered human scale fluid power applications.  Current rescue robots are electric.  They can navigate and observe, but do not have the needed force or power to perform rescue operations. Our goal is to develop a mobile fluid-power robot that can operate for a reasonable length of time (2 hours minimum), navigate in difficult terrain (urban disaster site), produce a required force (500 lbs of lift) with precision control and resulting dexterity (sufficient to apply medical test and treatment devices)  and transport a specified weight (250 lbs.).

Test Bed Role in Support of Strategic Plan

The Compact Rescue Robot occupies the power range from 100W to 1KW in the Center's efforts to apply to the full power range of applications.  This range is poorly addressed by fluid power today due to barriers, including a lack of compact power supplies, lack of miniature components and difficulty in tele-operation and control.

Description and Explanation of Research Approach

The existing applications at the human scale are simple one degree-of-freedom devices and generally dependent on large external power supplies.  Examples are log splitters and the "jaws of life" for extracting victims of accidents.  While the technology is very successful and indicates the potential of fluid power, their applications are limited.  Expansion to more degrees of freedom will require untethered power, miniaturized components and remote or autonomous operation.  Addressing these issues in the context of fluid power requires an imaginative leap into devices with this collection of requirements. Rescue in disaster scenarios is the leap we have taken.  Advances will be relevant to scenarios in the military, construction, agriculture, personal service and assistance to the handicapped and aged.  The state of the art in rescue robots has been reviewed by NIST in its periodic examination published in the Rescue Robotics Handbook. [1]  All entries are electrically powered, although a few extremely heavy ones have hydraulic manipulators attached.  Some have been exercised on a few disaster sites, but have not been capable of an actual rescue.  The military (DARPA) is pursuing rescue on the battlefield (BEAR robot [2] and legged field transportation (Big Dog [3], both employing hydraulics.  Neither would meet the specifications for TB4.

TB4, residing at the top of the three plane chart, will demand inputs from several projects to be successful.  Possible compact power supplies are a free piston engine compressor or pump, or a hot gas vane motor.  Safe and intuitive tele-operation will be accomplished through multi-modal haptic user interfaces.  The current incarnation of TB4 uses pneumatics, as H₂O₂ monopropellant producing 300 psi gas provides is the only source of power in an appropriate package for compact, untethered operation at this time. 

Achievements

In the past years, TB4 has advanced most through the development of two separate platforms.  At Vanderbilt, a four-legged crawler actuated by custom miniature high-pressure valves coupled with a Bimba cylinder and linear damper, has been designed and constructed (Figure 1).  The robot is controlled via CANbus communication to local microcontrollers at the three joints on each leg. In the past year, the Vanderbilt hardware has been pre-programmed with several low-level gaits for motion across relatively predictable surfaces, including a crawl, a walk, and a trot.  The Vanderbilt technology has been intended for use with hardware designed at Georgia Tech:  An operator workstation that uses two Sensable Phantom^TM haptic joysticks together with an A/V headset to provide feedback to the operator (Figure 2).  The workstation maps the two joysticks to the four legs of the robot, granting the operator intuitive control of gait and manipulation motions.  Georgia Tech has also developed a two-legged platform for manipulation testing and interim functionality.  These platforms are interfaced using xPC Target real-time software.

Quantifiable Performance Advantages:  A study, undertaken at Vanderbilt, used the mass and performance of the TB4 hardware in combination with properties of Center-developed power sources to point out the substantial improvements in energy efficiency that TB4 can bring to mobile human-scale platforms capable of significant manipulation.  These studies, shown in Table 1, demonstrate that using fluid-power can greatly reduced the mass of the system, especially as higher and higher run-times are expected.  This reduction in weight in turn allows the system to carry larger loads and last for longer periods of time on less energy, thereby validating many of the efforts of TB4 and associate CCEFP projects.

Hardware Advances:  In the past year, the Vanderbilt crawler has been completed, revised for functionality, documented, and brought to Georgia Tech.  Because it had originally been developed in a non-real time environment, changes were needed to ensure that the hardware functioned with the operator interface created at Georgia Tech.  An undergraduate researcher, Michael Baker, successfully converted several programs developed by Keith Wait at Vanderbilt from non-real time Simulink to xPC Target compatible Simulink.  He has thus far converted several of the key components needed for control of the motions, and is in the process of applying these to the pre-programmed gait software that had been developed at Vanderbilt.

Georgia Tech has also improved the two-legged testbed, which is used as simulation verification and as a platform for actuator control improvements.  Whereas the four-legged testbed couples a damper with a cylinder to make control of the position control joints simpler on a mechanical level, the two-legged testbed employs pressure sensors and Bimba^TM cylinders with position feedback.  This allows testing of alternate control strategies, such as passive control. In the last year, substantial improvements have been made to this platform.  Control was achieved via the operator workstation, using commands from the haptic joysticks to direct motion of the legs.  Electronics were reconfigured for a cleaner, more effective, and robust design.  The previous custom cylinders were replaced with Bimba hardware, as noted above, actuated by Festo proportional directional valves.  An undergraduate researcher, Michael Valente, redesigned the legs to accommodate the different, more compact cylinders.  He also enhanced the range of motion of the platform, using increased stroke length and improved design to substantially increase the range of motion of the legs, making them more capable of the desired lifting and motion tasks that the testbed aims to provide.  The revised design is also somewhat sleeker and lighter, yet maintains the kinematics used in all previous iterations of the robot (both at GT and at Vanderbilt).  This revised design is currently in construction and is expected to be complete by the end of the calendar year.

In the future, the Georgia Tech revised design will be completed and implemented with the new cylinders and improved range of motion.  This will be used to test control techniques targeted at precise movement of large loads by pneumatically actuated manipulators.

The Vanderbilt hardware will be completely integrated into the Georgia Tech platform, allowing usage of both the low-level, pre-programmed gaits and the semi-autonomous operator-guided gaits to control the robot.  Control techniques similar to the ones used on the two-legged Georgia Tech testbed will be implemented here, too.  The robot will also be further equipped with A/V feedback using a pan-and-tilt camera that moves together with operator motions of an associated headset, previously developed at Georgia Tech on the interim testbed.

Testing Environment:  While the low-level gaits used on the four-legged crawler have been tested in several outdoor environments, a necessary component to proving the versatility of the designed hardware is the usage of standardized "challenging" terrains.  Using the NIST [1] environments as a guide, a modular terrain block was created that can be configured to illustrate several difficult scenarios (Figure 3).

Figure 3: Robot negotiating the modular terrain environment at Georgia Tech

Future plans for this terrain include its use as a way of verifying the capabilities of the robot and simulation.

Advances in Simulation:  Another key component of TB4 is the hardware simulation.  The simulation was created in 2008/09, and uses an open source robotics library, courtesy of Seoul National University, known as SrLib.  This library lets the user select from a variety of joints and links to create kinematic representations of the desired hardware.  These are then placed in a simulated dynamic environment, where joints can be controlled either by actuated forces (representative of the actual hardware), or desired positions (representative of the ideal circumstance).  This serves several key functions:  First, it enables the testing of higher level control and operator interface features that would otherwise not be possible without a complete and functional robot, control scheme, and environment.  Similarly, it allows design of the operator interface in parallel with robot design, which can be tested within the safe and efficient bounds of the simulation.

A third feature of simulation is the result primarily of advances throughout the past year: it provides a better understanding of joint dynamics and allows simulated testing of new control techniques.  This is made possible by coupling the dynamic simulation of the robot with a low-level model of an actuator, consisting of the valve, cylinder, and associated controller.  This model, which has been discussed in two papers [4,5] published/accepted for publication this year, has been designed in Simulink and uses a simple proportional valve model, internal cylinder dynamics, and a friction model to generate a force output.  The model has been verified within Simulink to show near equivalent position and pressure behavior as physical systems, using a simple test setup as a measurable comparison.  These models have also been implemented together with the simulation, where they have demonstrated similar behavior and drawn conclusions on the effect of naturally occurring time delays in multi-platform simulations on the behavior of pneumatic models.

In the future, the dynamic actuator models will be applied to each of the joints and improved upon to ensure equivalence not only in single-platform simulations, but also when combining multiple software tools for a comprehensive dynamic simulation.  The model developed here will be used as a basis for advanced controls approaches, starting with establish pneumatic control techniques such as sliding mode control and LQR-derived control.  The simulation itself will continue to be used as a guide for interface design and operator control strategies.

Operator Interface and Robot Control:  The final key component of the TB4 platform is the operator interface.  This interface uses two Phantom haptic joysticks to control the legs of the robot, using a strategy known as the Follow-the-Leader gait to map the user to the robot for gait motions.  This strategy allows the user to place the front legs, while to computer decides where to place the rear ones based on knowledge of variables such as stability, safe footholds, and desired direction.  Several changes have been made in this interface in the past year.  Haptic guidance has been enabled, granting the user a better sense of telepresence through feedback from the joysticks.  The interface has also been redefined on a software level, using several modes of operation and internal state machines to provide clarity and ease of use to both the operator and the designer.  Several new gaits were added, including haptically guided ones developed at Georgia Tech and the pre-programmed low level gaits provided by Vanderbilt.

The operator interface has also benefited from a higher level controller developed at GT that places a penalty on stability (with respect to balance, not actuator performance) of the robot and relates it back to the user in the form of haptic feedback.  Thus, the user is guided to move in such a way that the stability of the robot is never compromised.  This operator-in-the-loop controller results in more effective overall motion without impeding too heavily on the user's level and sense of control.

Future plans for the operator interface are primarily focused on applying it to the four-legged crawler and ensuring complete functionality.  This entails coupling higher level control approaches that related robot balance and user desired motion with lower level actuator motion control to ensure that the user is able to effectively guide the robot across difficult terrain, as well as move the legs to lift items when necessary.

Education and Outreach:  TB4 has consistently provided an array of opportunities for impact and outreach, and the past year has been no different.  Because of its interactive set of components, TB4 is ideally situated to provide hands-on demonstrations to audiences from a wide range of backgrounds.  This past year, such demonstrations have been given to Atlanta city students visiting campus as part of National Robotics Week, FIRST students from across the country competing in the annual championship in downtown Atlanta, and visitors from a variety of other universities.  Additionally, the robot was featured as an example of the future of fluid power at a teaching enhancement session for Atlanta area FIRST students on fluid power, organized in conjunction with Georgia Tech's RoboJackets organization.

TB4 has also supported several undergraduate researchers, as noted throughout the summary of achievements.  This past summer, an REU, Allison Byrum, contributed towards control and dynamic modeling of the two-legged testbed.  In the fall, REU Michael Baker and undergraduate researcher Michael Valente both worked on TB4, integrating the Vanderbilt model with the Georgia Tech system, constructing terrain obstacles, and designing and constructing a revised manipulator design for the two-legged platform working with the newly acquired Bimba cylinders.

Finally, work on TB4 has resulted in several papers [4,5,6] on modeling, simulation, and interfaces of fluid-powered technologies, presented or accepted to be presented at conferences both within and outside the fluid power community.

In the future, TB4 will continue to provide compelling demonstrations that benefit from advances in fluid power research and natural appeal among varied audiences.  Because of its broad range of components, it will keep serving as an optimal source of research experiences for undergraduates and graduates alike, and will continue to result in publications across the industry.

References

1)    Messina, E., Jacoff, A., Scholtz, J., Schlenoff, C., Huang, H., Lytle, A. and Blitch, J., 2005, “Statement of Requirements for Urban Search and Rescue Robot Performance Standards,” Technical Report, Preliminary Report, National Institute of Standards and Technology.
2)    Murray, C.J., Robotic lifesaver [battlefield extraction assist robot]. Design News, 2006. 61(Copyright 2007, The Institution of Engineering and Technology): p. 46-50.
3)    Raibert, M., Blankespoor, K., Nelson, G., Playter, R., and the Big-Dog team, “Bigdog, the rough-terrain quadruped robot,” Proceedings of the 17th World Congress, The International Federation of Automatic Control, 2008.
4)    Book, W., Daepp, H., Kim, T. and Radecki, P., “An interactive simulation for a fluid-powered legged search and rescue robot,” Proceedings of 2010 International Symposium on Flexible Automation, Tokyo, July 2010.
5)    Daepp, H., Book W, “Modeling and Simulation of a Pneumatically Actuated Rescue Robot,” To be published at the International Fluid Power Expo, Las Vegas, Nevada, March 2011.
6)      Daepp, H., Book W, “An Interactive Simulation for a Fluid-Powered Search and Rescue Robot,” Proceedings of the 6th Annual Fluid Power Network International PhD Symposium, West Lafayette, Indiana, June 2010.