Marsette Vona: MSim-ATHLETE Info

 
MSim-ATHLETE example
download | browse source | javadoc

Mixed Real/Virtual Articulated Robot Simulator for ATHLETE

This is a special edition of MSim for NASA/JPL's ATHLETE (video introduction), including a non-proprietary graphical model of the robot. All of the documentation for and requirements of MSim also apply to MSim-ATHLETE.

The All-Terrain Hex-Limbed Extra-Terrestrial Explorer is a research platform under development at JPL for potential use in aiding humans in future Lunar missions. It incorporates a relatively large number of articulated Degrees-of-Freedom (DoF): each of its six limbs has six revolute joints, and in some contexts the wheel joints can also be kinematic, for a total of up to 42 DoF. Operating such systems (i.e. specifying of tasks and motions, nominally from a human operator, but potentially also from higher-level software) is challenging in part due to the relatively high dimension of the configuration space. Higher levels of autonomy—such as autonomous locomotion path planning and execution—are one answer, but there are still many cases where finer-grained control remains desirable provided that the potential tedium of specifying complex motions involving many joints can be reduced.

To this end, we are developing a theory and implementation of virtual articulation as a structured and natural interface that provides a richer semantics for the specification and communication of coordinated motions in high DoF systems. An operator adds virtual joints, and also possibly virtual links, to a kinematic simulation of the mechanism as a means of (1) constraining motion and (2) defining quantitative and concrete task-relevant DoF that need not correspond to any joint or effector in the actual mechanism. The mixed real/virtual kinematic model can then be interactively dragged and otherwise manipulated by the operator while the system maintains all joint constraints. Some hardware experiments and simulation examples are given below, most of which would be difficult to achieve without extensive special-case hand coding in existing systems for ATHLETE operations.

Hardware Experiments

These examples were recorded at NASA/JPL in January 2009. Hardware videos Copyright 2009 California Institute of Technology. Government sponsorship acknowledged.

Object Inspection

flv|ogv
Get the Flash Player to see this player.

In this experiment the operator designs a motion where a limb-mounted camera inspects a roughly spherical object while maintaining a constant distance. The operator directly models this constraint using a virtual spherical joint connecting the object (itself represented as a virtual link) and the camera. A secondary goal is to extend the space of reachable viewpoints by using the five other limbs to lean the hexagonal deck, but because the deck often carries a payload, we need to maintain its orientation. This is expressed by a virtual Cartesian-3 joint connected between the deck and the world frame. See below for the corresponding MSim interface view. After configuring the virtual articulations the operator can drag the camera with the mouse to scan the object. In this experiment we also used our direct-manipulation TRACK/ADMIN hardware interface to pose the limb holding the inspection camera, as shown next. An overhead crane serves as a safety backup in this experiment. Video speed 6x.

Object Inspection with TRACK/ADMIN

flv|ogv
Get the Flash Player to see this player.

For the object inspection experiment above we demonstrated integration of our direct-manipulation TRACK/ADMIN hardware interface to pose the limb holding the inspection camera. Video speed 1x.

Bimanual Pinch

flv|ogv
Get the Flash Player to see this player.

Two limbs execute a pinching maneuver with the pinch distance and angles controlled by virtual prismatic and revolute joints. See below for the corresponding MSim interface view. To safely grasp a rigid object some force control would also need to be incorporated. For this experiment the hex deck was partially supported by an overhead crane as simultaneously raising two limbs is not directly supported on the current hardware. Video speed 4x.

Trenching

flv|ogv
Get the Flash Player to see this player.

A trenching motion, with the hex deck leaning to extend reach. The motion is constrained and parameterized by several virutal prismatic and revolute joints, see below for the corresponding MSim interface view. A virtual Cartesian-3 joint, which allows 3DoF translation but no rotation, connected between the hex deck and the ground link maintains the deck orientation. Reachable trenching distance is increased by over 100% relative to what would be possible if the deck was held fixed. For safe digging some force control would also need to be incorporated. An overhead crane serves as a safety backup in this experiment. Video speed 10x.

Deck Camera Pan and Tilt

flv|ogv
Get the Flash Player to see this player.

Panning and tilting a camera built into the near side of the hex deck. The camera does not have its own pan and tilt actuators, but can still be aimed using postural motions. To operate this motion we used an assembly of two virtual revolute joints parameterizing the pan and tilt. See below for the corresponding MSim interface view. An overhead crane serves as a safety backup in this experiment. Video speed 10x.

Simulation Examples

These examples were recorded using a detailed graphical model of the robot developed by the RSVP team at NASA/JPL, not the non-proprietary model distributed with the MSim special edition available on this page. Other than that there are no significant differences between the two editions.

Object Inspection

flv|ogv
Get the Flash Player to see this player.

Five wheels are locked to the ground link and the remaining wheel is given an initial pose that aims its tool camera at a sphere reresenting an object of interest. A virtual spherical joint at the center of the object constrains the relative pose of the camera, which may then be dragged to inspect the object from different viewpoints. See above for the corresponding hardware experiment

Bimanual Pinching

flv|ogv
Get the Flash Player to see this player.

A virtual link o is added to represent a manipulated object, and serial chains of one prismatic and one revolute joint are added between o and the foot links on two adjacent ATHLETE limbs. The geometry of the grasp can be configured by manipulating these virtual joints, and the pinched object can be positioned in space by dragging o. See above for the corresponding hardware experiment.

Radial Trenching

flv|ogv
Get the Flash Player to see this player.

A trenching motion, with the hex deck leaning to extend reach. The motion is constrained and parameterized by several virutal prismatic and revolute joints. See above for the corresponding hardware experiment. A virtual Cartesian-3 joint, which allows 3DoF translation but no rotation, connected between the hex deck and the ground link maintains the deck orientation. Reachable trenching distance is increased by over 100% relative to what would be possible if the deck was held fixed.

Deck Camera Pan and Tilt

flv|ogv
Get the Flash Player to see this player.

Two virtual revolute joints are added between the deck and ground link, constraining the deck posture to a 2D space of poses that pan and tilt one of the side-mounted cameras, which does not have dedicated pan/tilt actuators. The added joints also give a concrente parameterization so that a particular pan/tilt scan sequence can be commanded. See above for the corresponding hardware experiment.

Deck Posture

flv|ogv
Get the Flash Player to see this player.

With the wheels locked relative to the ground link, the deck posture may be interactively manipulated in orientation and translation. The current version of MSim-ATHLETE does not yet compute center-of-mass or joint torque related qantities, so candidate motions developed in MSim-ATHLETE need to be verified for static stability and torque feasibility before uplink.

Crew Module Inspection

flv|ogv
Get the Flash Player to see this player.

One limb is posed to aim its tool camera at the surface of a cylinder representing a crew habitat, and a revolute-prismatic virtual joint chain is added to constrain the camera to a 2D space of poses in a concentric cylindrical patch. The camera view axis is effectively constrained by these virtual joints to stay normal to the crew module surface, and either joint may be locked to permit rotation or translation only. The camera can be moved either by operating the virtual joints or by dragging it directly. The virtual joints define a specific parameterization for the camera motion.

Postural Reaching

flv|ogv
Get the Flash Player to see this player.

Five wheels are locked to the ground link and the remaining wheel is interactively dragged in a reaching motion that might represent trenching or rock pushing. At first, the hexagonal deck is also locked, limiting the limb motion. The deck is then unlocked to permit extended reach. Finally, a virtual Cartesian-3 joint, which allows three orthogonal translations and no rotation, is interactively added between the deck and ground links to prevent the deck from twisting but still allowing some reach extension.

Download

MSim-ATHLETE is released in precompiled source form under the GNU GPL.

A lighter-weight jar is also available which includes MSim-ATHLETE, MSim, and VonaUtils but no third-party dependencies. There is also an extra-light jar which contains MSim-ATHLETE only.

Related Work

We are also devloping a hardware device called TRACK/ADMIN for master-slave teleoperation of the ATHLETE limb [MNPTMV08], and a group at Stanford university has considered the problem of autonomous kinodynamic path planning for ATHLETE [HBLW06]. The MSim related work also applies.

MNPTMV08David S. Mittman, Jeffrey S. Norris, Mark W. Powell, Recaredo J. Torres, Christopher McQuin, Marsette A. Vona. Lessons Learned from All-Terrain Hex-Limbed Extra-Terrestrial Explorer Robot Field Test Operations at Moses Lake Sand Dunes, Washington. AIAA SPACE Conference, 2008.
HBLW06K. Hauser, T. Bretl, J.C. Latombe, and B. Wilcox. Motion Planning for a Six-Legged Lunar Robot. Workshop on the Algorithmic Foundations of Robotics (WAFR), 2006.

Acknowledgements

Jeff Norris, David Mittman, and other collaborators at JPL facilitated this work by providing detailed info about the ATHLETE hardware and existing software.

This project was funded in part under the NASA/JPL SURP and NSF EFRI programs.

ATHLETE VRML model in demo animations courtesy of RSVP Team NASA/JPL/Caltech (model is proprietary and not included in download).

The MSim acknowledgements also apply.

Disclaimer

     THIS INFORMATION AND/OR SOFTWARE IS PROVIDED BY THE AUTHOR "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS INFORMATION AND/OR SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 

vona@mit.edu © 2009 Marsette Vona Tue May 5 20:23:26 EDT 2009