Jump to Content
Guides
v0.0
Log In
Guides
Log In
Moon (Dark Mode)
Sun (Light Mode)
v0.0
Guides
Set Up Object Manipulation Benchmark Tasks and Evaluation Measures
Search
Overview
Welcome to the Thousand Brains Project Documentation!
Vision of the Thousand Brains Project
Long-Term Goals and Principles
Short-Term Goals
Challenging Preconceptions
Capabilities of the System
Architecture Overview
Sensor Modules
Learning Modules
Cortical Messaging Protocol
Other Aspects
Bringing it Together
Application Criteria & Current Capabilities
Benchmark Experiments
Results from Alternative Implementations
FAQ - Thousand Brains Project
Glossary
Further Reading
How to Use Monty
Getting Started
Code Base Structure
Running Benchmarks
Logging and Analysis
Customizing Monty
Tutorials
Running Your First Experiment
Pretraining a Model
Running Inference with a Pretrained Model
Unsupervised Continual Learning
Multiple Learning Modules
Common Issues (And how to Fix Them)
How Monty Works
Implementation Overview
Experiment
Model
Environment & Agent
Observations, Transforms & Sensor Modules
Policy
How Learning Modules Work
Learning Module Outputs
Evidence Based Learning Module
Evidence as a Similarity Measure
Connecting LMs into a Heterarchy
FAQ - Monty
Open Questions
Contributing
Why Contribute?
Ways to Contribute
Documentation
Contributing Tutorials
Comment on Issues
Request For Comments (RFC)
Pull Requests
Contributor License Agreement
Pull Request Flow
Reviewing a Pull Request
Ways to Contribute to Code
Identify an issue to work on
Style Guide
Guides for Maintainers
Triage
Community
Code of Conduct
Governance
Our Licence and Patent Pledge Explained
TBP Based Papers
Project Showcase
Future Work
Project Roadmap
Sensor Module Improvements
Extract Better Features
Detect Local and Global Flow
Learning Module Improvements
Use Off-Object Observations
Reinitialize Hypotheses When Starting to Recognize a new Object
Improve Bounded Evidence Performance
Use Models with Less Points
Make it Possible to Store Multiple Feature Maps on one Graph
Test Particle-Filter-Like Resampling of Hypothesis Space
Re-Anchor Hypotheses
Less Dependency on First Observation
Deal with Incomplete Models
Implement & Test GNNs to Model Object Behaviors & States
Deal with Moving Objects
Motor System Improvements
Use a Different Policy for Learning than for Inference
Bottom-Up Exploration Policy for Surface Agent
Top-Down Exploration Policy
Bottom-Up Policies for Distant Agent
Calculate Saliency Map in View Finder and use it to Saccade to Region of Interest
Learn Policy using RL
Decompose Goals into Subgoals & Communicate
Interprete Goal States in Motor System & Switch Policies
Voting Improvements
Use Pose for Voting
Outline Routing Protocol/Attention
Generalize Voting to Associative Connections
Can we change the CMP to use displacements instead of locations?
CMP & Hierarchy Improvements
Figure out Performance Measure and Supervision in Heterarchy
Add Top-Down Connections
Add Associative Connections
Run & Analyze Experiments with >2LMs in Heterarchy Testbed
Run & Analyze Experiments in Multiobject Environment Looking at Scene Graphs
Test Learning at Different Speeds Depending on Level in Hierarchy
Send Similarity Encoding Object ID to Next Level & Test
Environment Improvements
Make Dataset to Test Compositional Objects
Set up Environment that Allows for Object Manipulation
Set Up Object Manipulation Benchmark Tasks and Evaluation Measures
Create Dataset and Metrics to Evaluate Categories and Generalization
Create Dataset and Metrics to Test new Feature-Morphology Pairs
Framework Improvements
Add Infrastructure for Multiple Agents that Move Independently
Automate Benchmark Experiments & Analysis
Add More Wandb Logging for Learning Unsupervised
Add GPU Support for Monty
Use State Class Inside of LMs
Make Configs Easier to Use
Find Faster Alternative to KDTree Search
OSS/Communication Improvements
Organize & Start Podcast Series
Make More Condensed Videos About the Project & Monty
Powered by
Set Up Object Manipulation Benchmark Tasks and Evaluation Measures
Updated 24 days ago
