How shalikadulaj/multi-agent-ai-hybrid-truck works
This is a research prototype for energy management of an assumed P4 heavy-duty hybrid truck. Its primary workflow is CLI invocation → tabular Q-learning training or saved-policy evaluation → held-out mission benchmarking and CSV/model artifacts. A separate, hand-coded multi-agent simulation provides supervisory, powertrain, battery, driver-assistance, and performance layers; README distinguishes its legacy heuristic from the learned Q policy. The sampled excerpts support only selected internal calls, so documented but unsampled capabilities are included with README-backed relationships.
Main entry points: rl_training.py Training workflow · agents.py Agent orchestrator · main.py CLI entry · artifacts Policy and results
Components 20
Experiment entry
- CLI entry
src/multi_agent_ai/main.py
RL experiment
- Training workflow
src/multi_agent_ai/rl_training.py - Duty-cycle generation
src/multi_agent_ai/rl_training.py - P4 plant transition
src/multi_agent_ai/rl_training.py - Q-table learning
src/multi_agent_ai/rl_training.py - Learned EMS policy
src/multi_agent_ai/rl_training.py - Rule baseline
src/multi_agent_ai/rl_training.py - Engine-only baseline
src/multi_agent_ai/rl_training.py
Multi-agent control
- Agent orchestrator
src/multi_agent_ai/agents.py - Mission supervisor
src/multi_agent_ai/agents.py - Powertrain policy
src/multi_agent_ai/agents.py - Pedal demand model
src/multi_agent_ai/agents.py - Battery health limits
src/multi_agent_ai/agents.py - Driver assistance
src/multi_agent_ai/agents.py - Performance layer
src/multi_agent_ai/agents.py - Sustainability reward
src/multi_agent_ai/agents.py
Evaluation and outputs
- Held-out evaluation
src/multi_agent_ai/rl_training.py - Benchmark metrics
src/multi_agent_ai/rl_training.py - Policy and results
artifacts
Other components
- Researcher
Connections 22
- Researcher → CLI entry: invokes
README.md - CLI entry → Training workflow: delegates
src/multi_agent_ai/main.py - Training workflow → Duty-cycle generation: generates routes
src/multi_agent_ai/rl_training.py - Training workflow → Q-table learning: trains
src/multi_agent_ai/rl_training.py - Q-table learning → P4 plant transition: steps environment
src/multi_agent_ai/rl_training.py - Q-table learning → Duty-cycle generation: uses cycles
src/multi_agent_ai/rl_training.py - Learned EMS policy → P4 plant transition: selects actions
src/multi_agent_ai/rl_training.py - Held-out evaluation → Learned EMS policy: evaluates
src/multi_agent_ai/rl_training.py - Held-out evaluation → Rule baseline: compares
README.md - Held-out evaluation → Engine-only baseline: compares
README.md - Held-out evaluation → Benchmark metrics: reports metrics
README.md - Training workflow → Held-out evaluation: benchmarks
README.md - Training workflow → Policy and results: writes outputs
README.md - Agent orchestrator → Mission supervisor: coordinates
src/multi_agent_ai/agents.py - Powertrain policy → Pedal demand model: computes demand
src/multi_agent_ai/agents.py - Powertrain policy → Battery health limits: requests limits
src/multi_agent_ai/agents.py - Agent orchestrator → Powertrain policy: coordinates
src/multi_agent_ai/agents.py - Agent orchestrator → Driver assistance: coordinates
README.md - Agent orchestrator → Performance layer: coordinates
src/multi_agent_ai/agents.py - Performance layer → Sustainability reward: evaluates reward
src/multi_agent_ai/agents.py - Mission supervisor → Battery health limits: sets safety objectives
README.md - Mission supervisor → Powertrain policy: sets mission objectives
README.md