Difference between revisions of "David Wilkie's Bio-Inspired Robot Project"

From GICL Wiki
Jump to: navigation, search
(Week 2)
(Introduction)
 
(4 intermediate revisions by one user not shown)
Line 3: Line 3:
 
This class is emphasizing the construction of comprehensive models for bio-inspired robots.  "Bio-inspired" means inspired by evolution's solutions to various problems (my definition).  For example, ant colonies use a [http://en.wikipedia.org/wiki/Stigmergy stigmergic] search strategy to find food: they walk in random paths; when one finds a source of food, it starts leaving a chemical path as it returns to the nest; this path compels other ants to follow, similar in result to the local beam search (which, in the words of Stuart Russel or Peter Norvig, says "Come over here, the grass is greener!").  This "meta-heuristic," inspired by nature, has been [http://scholar.google.com/scholar?q=ant+colony+optimization&hl=en&lr=&btnG=Search successfully applied to optimization problems].  In this class, our interest is robots inspired by nature's various methods of motion.
 
This class is emphasizing the construction of comprehensive models for bio-inspired robots.  "Bio-inspired" means inspired by evolution's solutions to various problems (my definition).  For example, ant colonies use a [http://en.wikipedia.org/wiki/Stigmergy stigmergic] search strategy to find food: they walk in random paths; when one finds a source of food, it starts leaving a chemical path as it returns to the nest; this path compels other ants to follow, similar in result to the local beam search (which, in the words of Stuart Russel or Peter Norvig, says "Come over here, the grass is greener!").  This "meta-heuristic," inspired by nature, has been [http://scholar.google.com/scholar?q=ant+colony+optimization&hl=en&lr=&btnG=Search successfully applied to optimization problems].  In this class, our interest is robots inspired by nature's various methods of motion.
  
For most students in the class, the initial model is,  somewhat paradoxically, the physical model.  That is, it is some lego or similar creation capable of undertaking some bio-inspired gait (wheels need not apply).  This is an ideal initial model due to the simplicity and versatility of legos.  Prototyping with legos can occur much quicker than in a CAD system (at for users with no experience with CAD).   
+
For most students in the class, the initial model is,  somewhat paradoxically, the physical model.  That is, it is some lego or similar creation capable of undertaking some bio-inspired gait (wheels need not apply).  This is an ideal initial model due to the simplicity and versatility of legos.  Prototyping with legos can occur much quicker than in a CAD system (at least for users with no experience with CAD).   
  
 
However, I've been working recently with Rich Primerano, who already has CAD models, kinematics simulation, and some physical sub-assemblies.  In keeping with my interests, I'll be working with him to develop comprehensive physics-based models, as well as behavioural, functional, system, and information models.
 
However, I've been working recently with Rich Primerano, who already has CAD models, kinematics simulation, and some physical sub-assemblies.  In keeping with my interests, I'll be working with him to develop comprehensive physics-based models, as well as behavioural, functional, system, and information models.
Line 14: Line 14:
 
== Week 2 ==
 
== Week 2 ==
 
I've returned to refactoring CobraCommander into something smaller and more manageable.  It uses the [http://www.ode.org/ Open Dynamics Engine] and [http://www.opengl.org/ OpenGL] to simulate a robotic system on terrain achieving some gait specified by angles or torques.  The advantage to this is the programmatic interface, which grants me greater control  over the simulation.
 
I've returned to refactoring CobraCommander into something smaller and more manageable.  It uses the [http://www.ode.org/ Open Dynamics Engine] and [http://www.opengl.org/ OpenGL] to simulate a robotic system on terrain achieving some gait specified by angles or torques.  The advantage to this is the programmatic interface, which grants me greater control  over the simulation.
I've started reviewing previous work in adstracting and simplifying CAD geometries for simulation.  While this is done in programs such as Ageia's PhysX, I haven't found any directly related research.  Preliminary sources and outline are [[here]]
+
I've started reviewing previous work in adstracting and simplifying CAD geometries for simulation.  While this is done in programs such as Ageia's PhysX, I haven't found any directly related research.  Preliminary sources and outline are at [[Simplified and Abstracted Geometry for Forward Dynamics]].

Latest revision as of 14:49, 9 October 2006

Introduction

This class is emphasizing the construction of comprehensive models for bio-inspired robots. "Bio-inspired" means inspired by evolution's solutions to various problems (my definition). For example, ant colonies use a stigmergic search strategy to find food: they walk in random paths; when one finds a source of food, it starts leaving a chemical path as it returns to the nest; this path compels other ants to follow, similar in result to the local beam search (which, in the words of Stuart Russel or Peter Norvig, says "Come over here, the grass is greener!"). This "meta-heuristic," inspired by nature, has been successfully applied to optimization problems. In this class, our interest is robots inspired by nature's various methods of motion.

For most students in the class, the initial model is, somewhat paradoxically, the physical model. That is, it is some lego or similar creation capable of undertaking some bio-inspired gait (wheels need not apply). This is an ideal initial model due to the simplicity and versatility of legos. Prototyping with legos can occur much quicker than in a CAD system (at least for users with no experience with CAD).

However, I've been working recently with Rich Primerano, who already has CAD models, kinematics simulation, and some physical sub-assemblies. In keeping with my interests, I'll be working with him to develop comprehensive physics-based models, as well as behavioural, functional, system, and information models.

Week 1

In an effort not to have to further unravel and refactor CobraCommander, my senior design project, I evaluated a couple dynamics simulation programs. First, I reviewed the 20-sim program, which I found referenced in a paper on an underactuated robotic walker, Dynamic Walking with Dribbel (PDF from IEEE). This program was used to tune the controller virtually, which was something Rich has been interested in doing. However, this program does not take geometry into account in the simulation nor does it support terrain.

I also checked out Microsoft's Robotics Studio, which is in free beta release. It's essentially a wrapper for the PhysX rigid-body dynamics engine. The product seems a bit too raw to use, but looks like it will eventually be useful. They also do CAD/Geometry simplification for simulation, something which Rich and I are studying.

Week 2

I've returned to refactoring CobraCommander into something smaller and more manageable. It uses the Open Dynamics Engine and OpenGL to simulate a robotic system on terrain achieving some gait specified by angles or torques. The advantage to this is the programmatic interface, which grants me greater control over the simulation. I've started reviewing previous work in adstracting and simplifying CAD geometries for simulation. While this is done in programs such as Ageia's PhysX, I haven't found any directly related research. Preliminary sources and outline are at Simplified and Abstracted Geometry for Forward Dynamics.