Andrew Danise Advisor: Prof. John Rieffel

Slides:



Advertisements
Similar presentations
Tuesday, May 14 Genetic Algorithms Handouts: Lecture Notes Question: when should there be an additional review session?
Advertisements

Tabu Search for Model Selection in Multiple Regression Zvi Drezner California State University Fullerton.
Understanding the Resistor Color Code
Final Project Presentation& Demo Zhi Dong Real Time FEM of Elasto-Plastic Simulation.
Recursion, Complexity, and Searching and Sorting By Andrew Zeng.
Evolving Scalable Soft Robots Senior Thesis Presentation Ben Berger Advisor: John Rieffel.
Lecture 8: 24/5/1435 Genetic Algorithms Lecturer/ Kawther Abas 363CS – Artificial Intelligence.
Evolving Virtual Creatures & Evolving 3D Morphology and Behavior by Competition Papers by Karl Sims Presented by Sarah Waziruddin.
GENETIC ALGORITHMS.  Genetic algorithms are a form of local search that use methods based on evolution to make small changes to a popula- tion of chromosomes.
HOT PLATE CONDUCTION NUMERICAL SOLVER AND VISUALIZER Kurt Hinkle and Ivan Yorgason.
Week 10 - Friday.  What did we talk about last time?  Graph representations  Adjacency matrix  Adjacency lists  Depth first search.
Houston, Wiebe & Batty. (2004) RLE Sparse Level Sets. RLE Sparse Level Sets Ben Houston Mark Wiebe Christopher Batty Fluids and Level Sets Session.
PROJECT TITLE Your name | Your teacher’s name | John W. Dodd Middle School.
screen shots Emma Jarman. Adding attachments What is an attachment? An attachment is an that has a file attached to it. The file could be.
Efficient Graph Traversal with Realistic Conditions by Olex Ponomarenko st Quarter Draft----
AP CSP: Pixelation – B&W/Color Images
CS1022 Computer Programming & Principles
Color Wheel/Color Theory
Genetic Algorithms.
Optimization Of Robot Motion Planning Using Genetic Algorithm
Learning Objectives To draw transformations of reflections, rotations, translations and combinations of these using graph paper, transparencies, and.
Fish Sticks.
Finite state machine optimization
Finite state machine optimization
University of New Mexico
Grid Pix Field Simulations and precision needed for a module
Rocket Difficulty Time Approximately 30–35 minutes Art Deco Clock
CHAPTER 15 Geometry Tina Rye Sloan To accompany Helping Children Learn Math10e, Reys et al. ©2012 John Wiley & Sons  
School of Computer Science & Engineering
For Monday Chapter 6 Homework: Chapter 3, exercise 7.
Lecture 22 Complexity and Reductions
Algorithm Analysis CSE 2011 Winter September 2018.
Extreme Graphical Simplification
Advanced Artificial Intelligence Evolutionary Search Algorithm
Controlling 3D Printers with Artificial Neural Networks
Search-Based Footstep Planning
Localizing the Delaunay Triangulation and its Parallel Implementation
Randomized Algorithms CS648
2008/12/03: Lecture 20 CMSC 104, Section 0101 John Y. Park
Computer Science 2 Hashing
Board: Objects, Arrays and Pedagogy
Academic Communication Lesson 3
Graphing and Describing “Rotations”
Level 0.
Tutorial 3 Empirical evaluation in AI
Edward de Bono’s Six Thinking Hats
Gears A gear is a rotating machine part, having cut like teeth, which fit together with another toothed part to transmit rotation. Geared devices can change.
Graphing and Describing “Rotations”
Functions.
Sorting "There's nothing in your head the sorting hat can't see. So try me on and I will tell you where you ought to be." -The Sorting Hat, Harry Potter.
Fish Sticks.
○ Hisashi Shimosaka (Doshisha University)
Directions: On each slide you will find fish of different colors, sizes, and numbers. You will also find bubbles for grouping the fish. On some slides.
Methods and Materials (cont.)
By Andrew Hilton Advisor: John Rieffel
CHAPTER I. of EVOLUTIONARY ROBOTICS Stefano Nolfi and Dario Floreano
Compact routing schemes with improved stretch
Introduction to Genetic Algorithm and Some Experience Sharing
More on HW 2 (due Jan 26) Again, it must be in Python 2.7.
More on HW 2 (due Jan 26) Again, it must be in Python 2.7.
Efficient Graph Traversal with Realistic Conditions
Traveling Salesman Problem by Genetic Algorithm
Computer Graphics Matrix Hierarchies / Animation
Scalable light field coding using weighted binary images
Robotics meet Computer Science
Patch Textures: Hardware Implementation of Mesh Colors
Simulation Fundamentals Static Stress/Shape Topology Tutorial
Lecture 22 Complexity and Reductions
Type your project title here Your name Your teacher’s name Your school
Mixed Reality: Are Two Hands Better Than One?
Presentation transcript:

Andrew Danise Advisor: Prof. John Rieffel Evolving Soft Robots Andrew Danise Advisor: Prof. John Rieffel

Why Soft Robots? Rieffel, Trimmer, Lipson, 2008. 1 minute Soft robots have no rigid structure, so they have more: - flexibility - maneuverability      - deformable shape and stretch Where could this be useful? - Squeeze in between rubble of Natural Disasters, search for survivors - Reach places that rigid body robots would have difficulty reaching Cons: - complexity of design Rieffel, Trimmer, Lipson, 2008. Whitesides et al., 2011.

How to Design Soft Robots? Rieffel et al., 2009. Rieffel and Smith, 2012. 1:30 minute To design a soft robot optimized for locomotion: Factors to consider: - What shape would be the best for movement? - What should be the mechanism it uses for movement? (oscillation of the stiffness of materials) - These two factors influence each other Lack of any analytical intuition: - use genetic algorithm : create an initial population of designs - create each design in a physics engine, and judge the fitness of the design in this case fitness is how far the design moves Get rid of the worst performing designs Generate new designs from the successful designs - To create new designs in the genetic algorithm you use encodings: Cheney et al., 2013.

Generative Encodings Lohn, Hornby, and Linden, 2004. Rieffel and Smith, 2012. 1 minute Ways to create new designs: - Direct Encoding:    - example: choose set of randomly placed tetrahedra and test their fitness    - Drawbacks:       - design not always viable (tetrahedra disjointed, overlapping)       - random (difficult to expand on a good design)       - essentially starting from scratch with each design - Generative Encoding:    - start from a single tetrahedra and “expand” it using a set of rules    - ex: tetrahedral face encoding grammar (explain the grammar in detail, use picture as a guide)    - Benefits:       - design is guaranteed to be feasible (if rules crafted well)       - easy to create widespread coordinated change (symmetry, modularity)         - ex: Table - Direct : increase the length of each leg individually - Generative : increase all the legs at the same time        Rieffel and Smith, 2012.

Sample Face Encoding Rules 1 minute Sample set of rules Discuss the specific rules that can be done (mention the colors are used instead of letters) Grow Subdivide relabel This slowly expand a soft robot design by applying the rules to the soft robot design, one face at a time

1st Expansion Applied Rule: Green → subdivide(Green, Blue, Orange, Red) 10 seconds Explain how the faces are expanded in order that they are added to the soft robot Rules applied one face at a time - get the next face - apply the rule corresponding with that faces color This is running in the simulation that I made

2nd Expansion Applied Rule: Blue → grow(Green, Blue, Blue) 10 seconds

3rd Expansion Applied Rule: Orange → relabel(Blue) 10 seconds

After 20 Expansions 30 seconds What’s the catch? Answer: The Stopping Problem

The Stopping Problem Rieffel, 2013. 1 minute The Main Drawback of generative encodings is the halting problem. Red line is the final generation. Black line is the fitness of the each generation Static Developmental Timings:  - expand using the generative encoding to a pre-decided number of generations Sometimes static developmental timing gets the best fitness possible (Top graphs) However, other times it does not get the best fitness possible (Bottom Graph) - would have achieved better fitness if it stop earlier or later Scaled Developmental Timings: - expand using generative encoding and slowly rachet up the number of generations This leads to research question -> next slide Rieffel, 2013.

Research Question How well can Scaled Developmental Timings help to address the stopping problem? How to make a soft robot move? 1 minute Due to the nature of Scaled Developmental Timings (slowly building up) it has many benefits over static developmental timings: For a given level of fitness achieved with Static and Scaled Developmental Timings: More computationally efficient (fewer tetrahedra to simulate) Less complex solutions (fewer tetrahedra in final design) See if the process of racheting up the number of generations can help address the halting problem BUT FIRST!!! Need to find a way for the robots to move

Using Vibration for Movement 2 minutes To look at the research question, need a way to move the robots Already have a way to design the shape of a robot : generative encodings Look at vibration as a movement mechanism Similar to research into movement of tensegrity robots Play video: Discuss the implementation How one cylinder rotates around the other How the other cylinder is attached to the soft robot, always in the same spot Cylinders don’t collide with anything but can influence the soft robot

Implementation 3 minutes These are all first generation – easier for me to take the video How to implement the encoding Currently, the simulation uses only static developmental timings Stats on runtime: Graphics ON: - Population = 6, Generations = 2, ~10 minutes - Population = 10, Generations = 5, ~43 minutes Sample Population = 40 (verify) Sample Generation = 1000 (verfiy)

Sample Generated Design 1 minute 3D printed robots: Simulated rendering of generated tetrahedral mesh Corresponding real 3D printed design Pager motors could be inserted into the printed design Work similarly to the simulated vibration Could be used to produce real soft robots that can move

Future Work Evolve and 3D print more designs and make them move Run simulation on the Union College cluster Compare Static and Scaled Developmental Timings 1 minute Explain what parts I have already done Explain the future work to do

Questions?