PhD dissertation: Bootstrapping Vehicles 2012-11-07

Art by striatic.

Andrea Censi. Bootstrapping Vehicles: a formal approach to unsupervised sensorimotor learning based on invariance. Technical Report, California Institute of Technology, 2012. pdf supp. material slidesbibtex

This is version 1.3 of my dissertation; I was not satisfied with the "official" 1.0 version, and I'm still adding material that I didn't have time to write. Because this is still a work in progress, any feedback is most appreciated.

Datasets

Media

Data used for calibration by correlation.

other videos
for Chapter 11

Vehicles navigation using bootstrapped models.

other videos
for Chapters 12-13

Larning BGDS model with streaming data

other videos
for Chapter 14+

Software

All software is available for download from various GitHub projects (I am working on a one-click installation for Amazon EC2; inquire if interested).

These are the main pieces:

  • boot12env is the "root" repository that contains scripts for setting up a virtual environment and checking out the other packages.
  • BootOlympics is the package responsible for interfacing agents and robots, loading/saving data and running the benchmarks, such as prediction, servoing, etc.
    • bvapps contains the configuration files for the simulations/experiments.
    • boot_agents contains the implementation of the agents (BDS, BGDS, DDS, etc.).
  • PyVehicles is used to run the Vehicles simulations.
  • PyGeometry implements all differential geometry functions.

These are miscellaneous utilities for creating reports, videos, and general plumbing: