Embodied AI Architect

A design environment for creating and evaluating autonomous agents, with hardware/software codesign space exploration and optimization.

Features

• Mission-Driven Workflow: Define a mission, qualify goals, select sensors/actuators, synthesize and analyze a complete system
• Sensor & Actuator Selection: Search, compare, budget, and fusion analysis for system components
• System Synthesis: Generate system architectures and bills of materials from mission specifications
• System Analysis: Power, latency, thermal, SWaP-C, and safety analysis of synthesized systems
• Model Analysis: Analyze PyTorch model structure and compute requirements
• Hardware Profiling: Recommendations for edge/cloud deployment
• Multi-Hardware Benchmarking: Local CPU, remote SSH, Kubernetes backends
• Interactive Chat: Claude-powered architect for design decisions
• SoC Optimization: LangGraph-based RTL optimization loop (experimental)

Mission Workflow Quick Start

The mission-driven workflow is the primary way to design an embodied AI system:

# 1. Create a mission
branes mission new vineyard-sprayer --goal "Autonomous vineyard spraying drone"

# 2. Qualify design goals and constraints
branes design qualify --mission vineyard-sprayer --auto

# 3. Select sensors and actuators
branes sensor search "stereo camera for VIO"
branes sensor select vineyard-sprayer visual.stereo_camera
branes actuator search "pump for spraying"
branes actuator select vineyard-sprayer fluid.sprayer

# 4. Analyze sensor budget and fusion
branes sensor budget vineyard-sprayer
branes sensor fusion vineyard-sprayer

# 5. Generate a design plan and synthesize the system
branes design plan --mission vineyard-sprayer --static
branes synthesize system vineyard-sprayer

# 6. Run system-level analysis
branes analyze-system power vineyard-sprayer
branes analyze-system swap vineyard-sprayer

See docs/quickstart-mission.md for a complete walkthrough.

Installation

Automated Setup (Recommended)

# Clone the repository
git clone https://github.com/branes-ai/embodied-ai-architect.git
cd embodied-ai-architect

# Run the setup script
./bin/setup-dev-env.sh

# Activate the environment
source .venv/bin/activate

# Verify installation
embodied-ai --help

Setup Options

# Full installation (Python deps + EDA tools)
./bin/setup-dev-env.sh --all

# Minimal installation (Python deps only)
./bin/setup-dev-env.sh --minimal

# SoC optimizer only (Python deps + EDA tools)
./bin/setup-dev-env.sh --soc

Manual Setup

If you prefer manual installation:

# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install core dependencies
pip install -e .

# Install with all optional dependencies
pip install -e ".[all,dev]"

# For SoC optimizer experiments, also install:
pip install langgraph langchain-anthropic

EDA Tools (for SoC Optimizer)

The SoC optimizer requires open-source EDA tools. The setup script installs OSS CAD Suite which includes:

• Yosys: RTL synthesis
• Verilator: Fast Verilog linting and simulation
• Icarus Verilog: Verilog simulation

Manual installation:

# Download OSS CAD Suite
wget https://github.com/YosysHQ/oss-cad-suite-build/releases/download/2025-12-12/oss-cad-suite-linux-x64-20251212.tgz
sudo tar -xzf oss-cad-suite-linux-x64-20251212.tgz -C /opt

# Add to PATH
export PATH="/opt/oss-cad-suite/bin:$PATH"

Usage

CLI Commands

# Show available commands
embodied-ai --help

# Analyze a PyTorch model
embodied-ai analyze model.pt

# Run full workflow
embodied-ai workflow run model.pt

# Benchmark on local CPU
embodied-ai benchmark model.pt --backend local

# Interactive chat session (requires ANTHROPIC_API_KEY)
export ANTHROPIC_API_KEY=your-key-here
embodied-ai chat

SoC Optimizer (Experimental)

The LangGraph-based SoC optimizer demonstrates agentic RTL optimization:

cd experiments/langgraph/soc_optimizer

# Run with mock mode (no LLM)
python workflow.py

# Run with Claude for RTL optimization
export ANTHROPIC_API_KEY=your-key-here
python workflow.py --with-llm

# Run simple loop (no LangGraph dependency)
python workflow.py --simple

Project Structure

embodied-ai-architect/
├── bin/                    # Setup scripts
│   └── setup-dev-env.sh    # Development environment setup
├── src/embodied_ai_architect/
│   ├── agents/             # Agent implementations
│   ├── cli/                # Click-based CLI
│   └── llm/                # LLM integration
├── experiments/
│   └── langgraph/          # LangGraph experiments
│       └── soc_optimizer/  # RTL optimization loop
├── prototypes/             # Research prototypes
│   ├── drone_perception/   # Real-time perception pipeline
│   └── multi_rate_framework/  # Zenoh-based multi-rate control
├── docs/                   # Documentation
└── tests/                  # Test suite

Development

# Run tests
pytest tests/

# Run single test
pytest tests/test_file.py::test_function -v

# Linting and formatting
black src/ tests/ --line-length 100
ruff check src/ tests/

Environment Variables

Variable	Description
ANTHROPIC_API_KEY	Required for Claude-powered features

Related Repositories

• embodied-schemas: Shared Pydantic schemas
• graphs: Analysis tools and roofline models
• systars: Amaranth HDL systolic array generator

License

MIT License - see LICENSE for details.