Release Notes: Smart Route Planning Agent#

Version 2026.3.0-rc1#

Release Date: November, 2026

New:

  • Added reasoning-based route planning using a configurable language model served by OpenVINO Model Server (OVMS), with deployment support in Docker Compose.

  • Added structured model responses and route validation, with automatic fallback to rule-based planning when the model is unavailable, times out, or returns an invalid decision.

  • Added Intel CPU, GPU, and NPU hardware information to the UI through Metrics Manager in Docker Compose deployments.

Improved:

  • Updated dependencies to address reported vulnerabilities and raised the minimum supported Python version to 3.11.

  • Pinned Gradio to a specific source commit and introduced multi-stage Docker builds that compile and package its frontend assets.

  • Added configurable reasoning request timeouts and OVMS KV-cache sizing, with persistent model caching in Docker Compose.

  • Updated the Docker Compose OVMS image to 2026.4.0 and increased the default reasoning timeout to 120 seconds to accommodate CPU inference with the documented Qwen models.

  • Updated documentation links and sparse-checkout instructions for the Metro AI Suite repository.

Deprecated:

  • Helm chart deployment is deprecated starting with this release. Use Docker Compose for new deployments and plan to migrate existing Helm deployments to Docker Compose.

Upgrade Notes:

  • Python 3.10 is no longer supported. Use Python 3.11 or later for source environments; the application container continues to use Python 3.12.

  • Update api_hosts entries in src/data/config.json to use host:port without the ws:// prefix.

  • To enable reasoning-based planning in Docker Compose, set REASONING_MODEL_NAME.

Version 2026.2.0#

Release Date: September 10, 2026

Improved:

  • Updated dependency specifications to use compatible-release (~=) versioning based on the PEP 440 standard.

  • Updated all dependencies to latest compatible versions.

  • Pinned uv installer to semantic versioning range (>=0.9.22,<1.0.0) for reproducible Docker builds.

  • Removed unused and a few transitive dependencies for a cleaner dependency tree.

Version 2026.1.0#

Release Date: May 14, 2026

New:

  • Deployment with Helm chart.

Improved:

  • Introduced a websockets client based connection to multiple Smart Traffic Intersection agents.

  • Replace the previous gradio based polling logic to asyncio based non-blocking calls at regular intervals.

Version 1.0.0#

Release Date: April 1, 2026

The Smart Route Planning Agent (SRPA) is a new sample application which presents the use of Hybrid AI in combination with the Smart Traffic Intersection Agent (STIA). SRPA acts as the orchestration agent, accepting live input from multiple Smart Traffic Intersection Agents and making routing decisions. It mimics a real-world deployment scenario, where an orchestration agent on the cloud communicates with edge agents to accomplish a particular goal. The goal for SRPA is to identify the most optimal route between two points using the live traffic data provided by the STIA, which mimics the agent deployed at the edge. A UI is provided to select two end points for the route, show recommended route options, and get information from edge agents.

New:

  • Real-time Traffic Analysis: Comprehensive directional traffic density monitoring with MQTT integration.

  • VLM Integration: Vision Language Model (VLM)-powered traffic scene analysis with sustained traffic detection.

  • Sliding Window Analysis: 15-second sliding window with 3-second sustained threshold for accurate traffic state detection.

  • Camera Image Management: Intelligent camera image retention and coordination between API and VLM services.

  • RESTful API: Complete HTTP API for traffic summaries, intersection monitoring, and VLM analysis retrieval.

  • Concurrency Control: Semaphore-based VLM worker management for optimal resource utilization.

    • Impact: Prevents VLM service overload and ensures reliable traffic analysis.

  • Image Retention Logic: Camera images persist with VLM analysis for consistent data correlation.

    • Impact: API responses show actual images analyzed by VLM, improving traceability and debugging.

  • Enhanced Error Handling: Comprehensive error management across MQTT, VLM, and image services.

    • Impact: Improved service reliability and diagnostic capabilities.

Known Issues:

  • Helm is not supported.

  • This release includes only limited testing on Standalone and Developer Node versions of Edge Microvisor Toolkit, some behaviors may not yet be fully validated across all scenarios.