physically-based modelling of robotic explorers employing th
TRANSCRIPT
PHYSICALLY-BASED MODELLING OF ROBOTIC EXPLORERS EMPLOYING THE EARBOT
NAVIGATIONAL CONTROL
Darin Rajan, Purdue University, IN/ Education Associates Program
Craig Slyfield, UC Berkley, CA/ Education Associates Program
Alexander Twombly (IC), USRS-RIACS, BioVIS Center, Moffett Field, CA
Jeffrey Smith (SLR), NASA-Ames Research Center, BioVIS Center, Moffett Field, CA
Richard Boyle (SLR), NASA-Ames Research Center, BioVIS Center, Moffett Field, CA
Introduction• EarBot is a biologically-inspired vestibular system that is
to be applied to the underlying control mechanisms of navigation, balance, and motor control found in robotic explorers, namely the SCORPION.
• The testing and development of these systems are done in simulation, where the walking SCORPION is in a physically emulated, true-to-life environment.
• This simulation is used to measure performance benchmarks by providing an overall integration platform of robotic systems, control strategies, and an assortment of terrain. Such benchmarks include:– Gyro and acceleration feedback– Omni-directional motion control feedback
Overview• The SCORPION is an eight-legged walking robot. The legs are
controlled by a basic motion pattern (BMP) generator to achieve walking behaviors and styles1.
• Simulating the SCORPION, and its control mechanisms, is done by the combination of two systems working jointly-together. First at the core of the simulation is NRG, a graphics wrapper simulation which interfaces a dynamic, physics-based engine created by Arachi, Inc. Second is 3D StudioMax, a 3D modeling tool created by Discreet, Inc.
• NRG is in charge of all the physics, viewing, and control in the simulation. Each system (both static and dynamic) that appears in the simulated environment is specified in a modeling system such as 3D Studio Max. Through NRG’s built-in plug-in to 3D StudioMax, models can be exported to a .XML format (.3ML) for viewing. The xml exporter is able to make most of the information available in the .max file description, including geometry, color, texture, hierarchy, cameras and lights. The models exported via the xml exporter may be used for either static display, kinematic animation, or dynamic interaction. Dynamic models require some additional information that is not normally associated with graphic models. This information include, link joints, object mass/inertia, friction properties, collision models used, sensors and control information.
• Finally, using a scripting language called Python, Arachi loads the system of models into the NRG viewer
1 Klaassen B., Linnemann R., Spenneberg D., Kirchner F. (2002) Biomimetic walking robot SCORPION: Control and modeling. Robotics and Autonomous Systems 41: 69-70Spenneberg D. & Kirchner F. (2001) An Approach Towards Autonomous Outdoor Walking Robots, Proceedings of 10th International Conference on Advanced Robotics (ICAR 2001), Budapest, Hungary.
SCORPION Simulation
SCORPION Simulation
NRG Executable Static and Dynamic Models
Arachi Physics Engine
Graphics Viewer Real World Physics
Controller
3D StudioMax
PolyTrans
SolidWorks
Python Script
SCORPION Anatomy
strutmotorB
motorC forearm
elbow
lowerarm
main
VestibularEach section of every leg is connected to a parent section. At the top of the hierarchy is the ‘main’ body. After physical properties have been added to the sections of the leg, a controller is interfaced through the physics engine via a Python script that is loaded at run-time of the simulation.
main
--strut
----motorB
------forearm
--------elbow
----------lowerarm
------motorC
--Vestibular Camera
SCORPION Hierarchy
SCORPION Navigation• A series of “Basic Motion Patterns” were
added to the SCORPION’s walking behavior. BMPs include fundamental walking algorithms, such as forwards, backwards, and sideways motion. Through a combination of keyboard and mouse callbacks the end user of the SCORPION simulation is able to navigate through the simulation.
• Each of these behaviors are added to the SCORPION at different levels of priority, depending on the frequency of the desired callback. That is if both forward and sideways BMPs are active and if sideways motion has a higher precedence, the SCORPION will walk more to the side than it does straight.
• The Vestibular camera allows the SCORPION to stabilize its view and create a fixation point. The camera is then interfaced with a Neural Network, where it will be able to gather information on any point in the simulation, focus on it, and finally accurately stabilize the camera’s view. A simple demo program, pictured, demonstrates this
Method of Simulation1. Create individual models of parts making up articulating objects
The first step in modeling a robotic explorer so that a simulation can be created is to develop a 3D model using SolidWorks. The CAD model contains the basic defining mesh for the rigid body structure.
2. Assembly of parts to create model
3. Conversion of assembly into file type readable by 3D StudioMax
4. Grouping and hierarchy
The SCORPION model is an exact replica of the original robot, assembled in 3D StudioMax. The entire model is divided into separate sub-objects that consist of struts, motors, joints, and legs. These objects fall into an ordered hierarchy of nodes and links.
5. Textures and lighting
Textures and lighting are needed to add realism to the simulation. It brings the digital, simulated world to life. This digitization is achieved by wrapping images around the geometries of the rover mesh.
6. Creating Terrains
Vertex translation of polygon mesh to create a crater.
Mars Texture
Brick Texture
Asphalt Texture
Using 3D StudioMax, meshes can be modeled around real life terrain features, which can test the rover’s agility and maneuverability in unique situations.
Model ComparisonsK9 214,384 polygons
SCORPION 550,298 polygons
MER 320,454 polygons
Direct Comparison of Model Data for Hip Joint
Simulated ModelTheoretical Model
Real Model
Current Status• The SCORPION model has been updated completely with the
a new high-res model and does not require the old model to be a ‘ghost’ or backbone to it.
• The controller is now able to guide the SCORPION to walk in all directions, and can be interfaced with the neural network to provide for Vestibular Camera stabilization.
• Many different types of terrains have been designed to challenge the rover and test the simulated movement produced, providing an initial test bed for the accurateness of robotic simulations.
• The walking pattern of the simulated SCORPION robot matches that of the original robot and allows for the sensors to be rea
Future Work
• Update the current high-res model with a higher resolution, more precise model to allow for a higher degree of accurateness.
• Update the user interface to allow for models to be loaded and terrains to be changed while the simulation is running.
• Upgrade the existing system to future versions of the Arachi Physics Engine.