Skip to content

Project Roadmap (Archived - Skills-First Approach)

⚠️ ARCHIVED: This document has been superseded by mvp_embodied_ai_platform.md.

Historical Context: This roadmap documented a skills-first implementation strategy where development would proceed through phases: build infrastructure → implement skills → add mission agent → hardware validation → advanced features. It represented a 6-month timeline (Oct 2025 - May 2026) with specific skill inventories and phase-based delivery.

Why Archived: The project pivoted to a capability-driven approach focused on end-to-end MVP delivery with concrete mission scenarios. The new MVP roadmap emphasizes transformer architecture exploration (LLM, VLM, VLA) and defines success through working missions rather than skill counts.

Valuable Content Preserved Here: - Detailed testing infrastructure requirements (pytest, CI, mock robot) - Comprehensive skill categorization (13+ skills across nav/perception/reporting/system) - Phase dependency mapping and risk analysis - Cloud agent collaboration velocity calculations - Hardware validation test suite specifications

Refer to this document when considering alternative implementation approaches or detailed testing/validation strategies.

Purpose (Original)

Define strategic milestones with specific, measurable deliverables that guide ShadowHound development from infrastructure to field-ready autonomy.

Last Updated: 2025-10-13 (Revised with specific deliverables)
Current Phase: Phase 0 (Infrastructure) - 90% Complete

Prerequisites

Phase Overview

📊 Progress Summary

Phase Status Completion Target Date
Phase 0: Infrastructure 🔄 In Progress 90% Oct 2025 ✅
Phase 1: Skills Foundation 🔜 Next 0% Nov 2025
Phase 2: Mission Agent 🔮 Planned 0% Dec 2025 - Jan 2026
Phase 3: Hardware Validation 🔮 Planned 0% Feb - Mar 2026
Phase 4: Advanced Autonomy 🔮 Future 0% Apr - May 2026

Phase 0: Infrastructure ✅ 90% Complete

Goal: Development environment and documentation infrastructure ready for rapid implementation

Duration: September - October 2025

Deliverables

✅ Completed

  • [x] Devcontainer with ROS2 Humble + DIMOS
  • [x] Package structure (3 packages: bringup, mission_agent, skills)
  • [x] Build system working (colcon build)
  • [x] Documentation infrastructure (MkDocs, GitHub Pages, Obsidian vault)
  • [x] Development workflows documented
  • [x] Cloud agent collaboration workflow
  • [x] Issue templates and PR templates
  • [x] Submodule protection (auto-formatting prevention)

🔄 Remaining (10%)

  • [ ] Testing infrastructure (pytest configuration, CI for tests)
  • [ ] Code quality checks in CI (black, flake8, mypy)
  • [ ] Mock robot interface for testing

Success Criteria

✅ Team can develop efficiently with clear docs and workflows
⏸️ Can run automated tests on every PR (blocked on testing infrastructure)

Next Action: Complete testing infrastructure → Phase 1

Phase 1: Skills Foundation 🔜 Target: November 2025

Goal: 10-15 working skills with comprehensive testing, validated in simulation

Duration: 2-3 weeks (with cloud agent assistance)

Deliverables

Core Infrastructure

  • [ ] RobotInterface implemented
  • Connects to DIMOS API
  • Wraps basic motion commands
  • Safety clamps (velocity limits, bounds checking)

  • Error handling and timeouts

  • [ ] SkillRegistry implemented

  • Skill base class with validation
  • Registration decorator (@register_skill)
  • Discovery mechanism

  • Execution engine with telemetry

  • [ ] Testing Infrastructure

  • pytest configuration
  • Mock robot interface
  • Test utilities and fixtures
  • CI integration (GitHub Actions)
  • [ ] nav.stop - Immediately stop all motion
  • [ ] nav.rotate - Rotate by angle (degrees)
  • [ ] nav.translate - Move forward/backward by distance
  • [ ] nav.goto - Navigate to (x, y, yaw) pose
  • [ ] nav.follow_path - Follow waypoint list

Perception Skills (3)

  • [ ] perception.snapshot - Capture camera image
  • [ ] perception.scan_lidar - Get lidar scan data
  • [ ] perception.detect_obstacles - Basic obstacle detection

Reporting Skills (3)

  • [ ] report.log - Log message with timestamp
  • [ ] report.speak - Text-to-speech (if available)
  • [ ] report.status - Report system status

System Skills (2-4)

  • [ ] system.health_check - Verify all systems operational
  • [ ] system.emergency_stop - Safe emergency shutdown
  • [ ] system.battery_status - Report battery level
  • [ ] system.reset - Reset to safe state (optional)

Success Criteria

  • ✅ All skills pass unit tests (>80% coverage)
  • ✅ Can execute "rotate 90 degrees, move forward 1 meter, stop" via skills API
  • ✅ Skills validated in simulation (Gazebo or mock)
  • ✅ Documentation complete for all skills

Estimated Effort: 40-60 hours (2-3 weeks with cloud agent for testing/docs)

Phase 2: Mission Agent 🔮 Target: December 2025 - January 2026

Goal: Natural language mission execution with real-time monitoring

Duration: 1-2 months

Deliverables

Mission Agent Core

  • [ ] DIMOS Agent Integration
  • OpenAI agent for single-shot commands
  • PlanningAgent for multi-step missions
  • Configuration switching (cloud vs local LLM)

  • Dependency: DIMOS Documentation Issue #7 - Integration guides

  • [ ] Mission Executor Node

  • ROS2 node wrapping mission executor
  • Skill execution via SkillRegistry
  • Status publishing
  • Error recovery

User Interface

  • [ ] Web Dashboard
  • Real-time camera feed
  • Mission status display
  • Performance metrics
  • Command input interface

  • Terminal/log viewer

  • [ ] Telemetry System

  • Mission execution tracking
  • Skill performance metrics
  • System health monitoring
  • Historical data storage

Integration

  • [ ] Camera Feed Integration
  • ROS2 topic subscription
  • Image compression/streaming

  • QoS configuration

  • [ ] Example Missions

  • 5+ documented example missions
  • Simple to complex progression
  • Test suite for mission validation

Success Criteria

  • ✅ Can execute "patrol the area" and robot follows multi-waypoint path
  • ✅ Web UI shows live camera feed and status
  • ✅ Mission success rate >90% in simulation
  • ✅ Response time <5s for multi-step commands

Estimated Effort: 80-120 hours (1-2 months, web UI via cloud agent)

Phase 3: Hardware Validation 🔮 Target: February - March 2026

Goal: Reliable operation on Unitree Go2 in controlled environment

Duration: 1-2 months

Deliverables

Hardware Integration

  • [ ] Skills on Real Hardware
  • All 13+ skills tested on Go2
  • Safety parameters calibrated
  • Performance benchmarks collected

  • Hardware-specific quirks documented

  • [ ] Safety Systems

  • Emergency stop validated
  • Collision avoidance working
  • Battery monitoring active
  • Failsafe behaviors tested

Testing & Validation

  • [ ] Field Test Suite
  • 10+ test missions designed
  • Indoor environment testing
  • Various terrain types

  • Edge case validation

  • [ ] Performance Metrics

  • Mission success rate >90%
  • Average response time <2s
  • Battery efficiency measured
  • Network reliability tested

Documentation

  • [ ] Operations Manual
  • Hardware setup guide
  • Safety procedures
  • Troubleshooting guide
  • Maintenance checklist

Success Criteria

  • ✅ 90%+ mission success rate in controlled indoor environment
  • ✅ Robot can run for 30+ minutes on single charge
  • ✅ Emergency stop works reliably
  • ✅ No safety incidents during testing

Estimated Effort: 60-80 hours (1-2 months of hardware testing)

Phase 4: Advanced Autonomy 🔮 Target: April - May 2026

Goal: Outdoor autonomy with vision-based navigation

Duration: 2-3 months

Deliverables

Vision & Perception

  • [ ] VLM Integration
  • Scene understanding via vision-language models
  • Object detection and tracking
  • Semantic scene parsing

  • Visual navigation assistance

  • [ ] Advanced Navigation

  • Dynamic obstacle avoidance
  • Terrain assessment
  • Weather/lighting adaptation
  • Path planning with vision

Multi-Robot (Optional)

  • [ ] Coordination
  • Multi-robot discovery
  • Task allocation
  • Collision avoidance
  • Fleet management

Persistence

  • [ ] World Model
  • Persistent map storage
  • Object memory
  • Location history
  • Learning from experience

Success Criteria

  • ✅ Can navigate outdoor environment autonomously
  • ✅ Adapts to changing conditions (lighting, weather)
  • ✅ Avoids dynamic obstacles reliably
  • ✅ Mission success rate >80% outdoors

Estimated Effort: 120-160 hours (2-3 months)

Dependencies & Risk Mitigation

Phase Dependencies

  • Phase 1 → Phase 2: Need working skills before mission agent
  • Phase 2 → Phase 3: Need working agent before hardware testing
  • Phase 3 → Phase 4: Need reliable hardware operation before advanced features

Known Risks

Phase 1 Risks

  • Risk: DIMOS API complexity
  • Mitigation: Start with simple skills, thorough mocking

Phase 2 Risks

  • Risk: LLM latency/reliability
  • Mitigation: Local LLM option, timeout handling

Phase 3 Risks

  • Risk: Hardware unavailability
  • Mitigation: Extensive simulation testing first

Phase 4 Risks

  • Risk: Outdoor complexity (weather, terrain)
  • Mitigation: Gradual outdoor exposure, fallback behaviors

Velocity Multipliers

Cloud Agent Usage

Target: Use cloud agents for 30-50% of implementation work

Phase 1 Candidates: - Testing infrastructure setup - Mock robot interface - Individual skill implementations (after first 2 as examples)

Phase 2 Candidates: - Web UI implementation (excellent cloud agent task) - Telemetry dashboard - Example mission collection

Phase 3 Candidates: - Test suite generation - Documentation updates - Performance analysis scripts

Expected Velocity Gains

  • Without cloud agents: 6-9 months total
  • With cloud agents: 4-6 months total (30-40% faster)

Validation

  • [x] Roadmap aligns with actual project state (Oct 2025 analysis)
  • [x] Each phase has specific, measurable deliverables
  • [x] Success criteria defined for each phase
  • [x] Dependencies and risks documented
  • [ ] Phase 1 tasks broken into GitHub issues
  • [ ] Cloud agent candidates identified for each phase

See Also

References