Multi-Robot Systems
CSCI 7000-006Monday, September 28, 2009
NikolausCorrell
Crafting a Research Project
• What is “research”?• Preliminary requirement: open question• Secondary: how to solve it• Hypothesis: states question and leads to
methodology• Sources of confusion
– You need to investigate what the questions are– You need to design your experiment– You need to optimize your system– You need to develop tools to investigate
Collaborative Lifting
• Problem: Lifting a box collaboratively
• Hypothesis: Problem can be encoded in a single cost function that allows gradient-based control
• Method: formal stability analysis
Gregory Brown
Collaborative Bouncing
• Problem: Bouncing a ball back and forth between two robots
• Hypothesis: Use a particle-filter for predicting system dynamics
• Method: Dynamical model and implementation
Mikael Ian Pryor
Probabilistic Patrolling
• Problem: Patrol an environment efficiently but unpredictable to the adversary
• Hypothesis: Use a balance between exploration and exploitation during coverage
• Method: Probabilistic algorithm, model, implementation
VijethRai
Probabilistic Localization with Geometric Constraints
• Problem: Localizing “intelligent” objects
• Hypothesis: Using the object geometry and simulated physics in a particle filterfor an RFID reader can improve localization accuracy
• Method: Particle filter combined with physics-based simulator
Neeti Shared Wagle
Reactive Coverage with Connectivity Constraints
• Problem: cover an environment while maintain connectivity
• Hypothesis: Constraints can be encoded in a global cost function
• Method: Stability analysis of gradient-based controller MaciejStachura
Probabilistic Path Generation for Data Ferrying in Unknown Sensor Deployments
• Problem: collecting data from sensor network using mobile robot
• Hypothesis: optimal planning always better or same than randomized even if node location is unknown
• Method: analysis and hardware validation
Anthony Carfang
Policy-space Learning of Tunable Locomotion Primitives
• Problem: learn to locomote unknown actuator configurations
• Hypothesis: The Natural Policy Gradient method can allow to find optimal policies in high-dimensional, continuous state space in real time
• Method: implementation in realistic simulation
Ben Pearre
Resource sharing in Multi-Robot Systems
• Problem: improve individual performance by relying on team sensors
• Hypothesis: Can Resource Sharing Make Up for Perception Deficiencies in a Multi-Robot Team?
• Method: Demonstration in real hardware
GPS
Peter Klein
Informed Flocking in Honey Bees
• Question: how do honeybees communicate the location of a new nesting site
• Hypothesis: Can the Robustness to Disturbances Shed Light into the Preferred Method of Informed Flocking in Honey-Bees?
• Approach: mathematical model and numerical simulation
Apratim Shaw
Mothership/DaughtershipCoverage Control Problem
• Question: how to best distribute capabilities in a system?
• Hypothesis: A hierarchical mothership(MS)/daughtership (DS) system can be applied to coverage control problems and is more efficient and scalable than a team of all MS or all DS.
• Method: mathematical model and numerical simulation Jason Durrie
An agent based approach to music generation
• Problem: generate nice music automatically
• Hypothesis: A threshold agent based model where each agent represents a note on the piano is capable of creating “good” sounding music.
• Approach: mathematical model and numerical simulation
Stephen Heck
MROS: Multi-Robot Operating System
• Problem: message passing in ROS limited to a single agent
• Hypothesis: broadcast message proxies can turn local message bus into message graph
• Implementation: Message proxy using BioNet
MarekSotola
Smart Sand
• Problem: Mapping hard to access environments
• Hypothesis: We can reconstruct the topology and sensing landscape of a cavity using large numbers of smart spheres that can establish their local position
• Method: implementation in ODE, analysis
Monish Prabhakar
Towards Truly Soft Robots
• Problem: Creating shape deformation and actuation from soft components
• Hypothesis: Given a soft smart sheet composed of cells that can be individuallyactuated and that can as a result actively change its shape, it is possible to createarbitrary 3D polygons by combining and contorting the 1D sheets in novel ways
• Method: Implementation of spring-mass model of actuator meshes in ODE
SwamyAnanthanarayan
Optimal plant placement
• Problem: place plants such that light and water are optimally used
• Hypothesis: Genetic algorithms will outperform gradient-based optimization in strongly-coupled, non-linear dynamic systems
• Method: Mathematical model, numerical simulation
Rhonda Hoenigman
Implementation
• Common resources/goals
– Manipulation
– Communication
– Mobile base
– ODE
– Matlab
• Create clusters and collaborate
Project report
• Motivation for your research
• Hypothesis
• Materials and Methods
• Results
• Discussion
• Conclusion
Scientific thesis in general
• Principally you need a hypothesis and write a dissertation to defend it
• The reality is often different
– Investigate interesting problem and variations
– Funding driven (not necessarily scientific)
– Change in direction/advising
• Solution: what is the most interesting question my material can answer? Drop all the rest.
This week
• Wednesday: Probabilistic Modeling
• Friday: Start course projects