CAMPD — API Documentation

Context-Aware Motion Planning Diffusion

CAMPD is a framework for training context-aware diffusion models on trajectory data for robot motion planning. It leverages classifier-free guidance and attention mechanisms to condition a U-Net diffusion model on structured, sensor-agnostic contextual information.

Note

For the project overview, paper, and demo videos, visit the project page.

Getting Started

Install the package:

pip install campd

API Reference

User Guide

• Core Concepts
  • Registry System
  • The Spec Pattern
  • Configuration and Attribute Propagation
• Launching Experiments
  • Via the campd-run CLI (Recommended)
  • Via Custom Python Script
• YAML Configuration Reference
  • Root Configuration Nodes
  • Example: Built-in train Structure
• Extending the Framework
  • Registering an Experiment
  • Registering Custom Network Architectures
  • Registering Custom Training Objectives
  • Registering Training Callbacks
  • Registering Inference Validators
• Troubleshooting
  • CUDA Graph Errors
  • KeyError: Unknown 'X' in registry 'Y'

Packages

• Models
  • Diffusion Model
  • Helpers
• Architectures
  • Temporal U-Net
  • Reverse Diffusion Base
  • Context Encoder
  • Layers
  • Attention
  • Layer Utilities
• Data
  • Trajectory Context
  • Embedded Context
  • Trajectory Dataset
  • Trajectory Sample
  • Normalization
• Training
  • Trainer
  • Training Objectives
  • Losses
  • Callbacks
  • Summary
• Experiments
  • Base Experiment
  • Training Experiment
  • Inference Experiment
  • Validators
  • Runner
• Utilities
  • Configuration
  • File Utilities
  • Torch Utilities
  • Random Utilities
  • Statistics
• Registry
  • System Overview
  • Available Component Registries