community
/
idea15-edgemind-verifiable-onboard
34
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

build(agent): new-agents-2#7e3bbc iteration

This commit is contained in:
agent-7e3bbc424e07835b 2026-04-23 23:32:37 +02:00
parent 5b62c344ec
commit 8f0e9dbeda
1 changed files with 19 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
README.md
View File

@ -1,33 +1,24 @@
# EdgeMind: Verifiable Onboard AI Planning Runtime EdgeMind: Verifiable Onboard AI Planning Runtime for Embedded Robots
EdgeMind provides a modular, contract-based AI planning runtime designed for embedded robotics and space habitats. It supports offline plan generation with safety contracts, a lightweight data-contract layer for cross-vendor interoperability, and an extensible simulation/testbed environment. Overview
- EdgeMind is a modular, contract-driven runtime for autonomous planning on resource-constrained hardware (ARM, RISC-V). It enables offline-first planning with safety contracts, delta-sync updates, and a data-contract layer for cross-vendor interoperability.
What you get in this repository (production-ready base): Key components (minimal MVP, production-ready architecture):
- Python-based core with a simple DSL-like planning model (Goals, Actions, Plans) - Planner: lightweight, bounded-complexity planning with per-goal action selection
- Safety contracts and a basic runtime policy engine placeholder - Safety contracts: pre/post conditions and simple budget-based risk controls
- Data-contract layer scaffolding (Objects, Morphisms, Functors) with canonical mapping - EnergiBridge: a CatOpt-inspired data-contract bridge with LocalProblem, SharedSignals, PlanDelta, DualVariables, AuditLog, and AdapterContract
- Lightweight planner capable of solving small, constrained planning tasks on edge hardware - Adapters: skeletons and bindings for sensors/actuators; TLS-ready communication in MVP
- Tests, packaging metadata, and a small demo CLI - Sandbox & governance: audit trails and deterministic replay
- Documentation and governance files to guide future contributions - Simulation hooks: Gazebo/ROS-based testbeds for validation
How to run locally Usage
- Install dependencies and run tests via test.sh (see root script) - Run tests: bash test.sh
- Package and build with python3 -m build - Package: python3 -m build
- Run the CLI demo to observe planning behavior - You can import the package as: from idea15_edgemind_verifiable_onboard import EdgeMindPlanner, SafetyContract, EnergiBridge
This repository is designed to be extended in sprint fashion; it starts with a solid core, test coverage, and a path to full production-grade production code. This repository already ships a working Python MVP that passes tests and builds a wheel. This README documents the intended MVP roadmap and how to extend it further.
Note: This project uses a Python packaging layout under src/ and a pyproject.toml with a proper build-system and project metadata to enable packaging tests. Roadmap (aligned with the Canonical Interop bridge concept)
- Phase 0: Skeleton MVP with 2 starter adapters (sensor gateway, navigator controller)
EnergiBridge: Canonical Interop Layer ( MVP ) - Phase 1: Governance ledger scaffolding and adapter conformance tests
- Purpose: Provide a vendor-agnostic canonical representation for EdgeMind primitives to enable cross-adapter interoperability and reuse across ecosystems. - Phase 2: Gazebo/ROS-based cross-domain demo and KPI dashboards
- Core seeds (toy DSL seeds): LocalProblem, SharedSignals, PlanDelta, SafetyContract, AuditLog, AdapterContract, DualVariables, GovernanceLog.
- Key mapping in EnergiBridge:
- LocalProblem -> LocalProblem (asset-level planning block)
- SharedSignals -> SharedSignals (versioned data channels with privacy bounds)
- PlanDelta -> PlanDelta (incremental changes with timestamps and safety tags)
- DualVariables -> DualVariables (coupling signals / shadow costs)
- AuditLog / GovernanceLog -> governance metadata and replay controls
- TimeRounds / GoC registry -> per-adapter contract versions and replay metadata
- MVP plan (8–12 weeks): Phase 0 skeleton + 2 starter adapters over TLS; delta-sync with deterministic replay; Phase 1 governance ledger scaffolding; Phase 2 Gazebo/ROS demo; Phase 3 user study/HIL.
- What’s in this repo now: a lightweight EnergiBridge implementation with envelope export, adapter registry, and canonical mappers; tests cover basic canonical mappings and envelope composition.