Skip to content

0.5.1

Latest
Latest

Choose a tag to compare

View all tags
@aleph-ra aleph-ra released this 16 Feb 10:37
· 247 commits to main since this release
0.5.1
d02a327

EmbodiedAgents v0.5.1: Expanded Vision Actions & Router Flexibility

We are releasing EmbodiedAgents v0.5.1, a patch release that brings significant quality-of-life improvements to the Vision component and increases the flexibility of the Semantic Router. This update focuses on enhancing the operational capabilities of agents, allowing them to autonomously capture data and execute more complex routing logic.

🌟 New Features

📸 Native Vision Actions

The Vision component has been upgraded with built-in actions, making it easier for agents to control their own visual data collection pipeline without external triggering logic. These actions can be triggered with events or used in the router.

  • record_video Action: Agents can now trigger video recording on specific image input topics for a set duration.
  • take_picture Action: Added support for instantaneous snapshot capture.
  • Reliability: We refactored action methods within the Vision component, preventing inference starvation during high-load operations.

🔀 Enhanced Semantic Routing

The SemanticRouter is now more flexible, allowing for broader control flow manipulation. One can route data to multiple topics in the single route. Or execute one or multiple Action(s) in a route.

  • List Handling: Routes can now handle lists of topics and actions, allowing a single semantic decision to trigger multiple downstream behaviors or broadcast to multiple topics simultaneously.
  • Arbitrary Action Execution: The router now supports the execution of arbitrary actions, further decoupling decision-making from hardcoded paths.

Learn more about Actions here.

🛠️ Enhancements & Improvements

  • Ollama "Think" Support: Added a think option toggle for Ollama models, enabling better integration with reasoning models (e.g., Qwen family) that utilize chain-of-thought outputs.
  • LLM/VLM Configuration: Increased the default max_new_tokens parameter for LLM and VLM configs to support more verbose model outputs by default.
  • Documentation:
  • Updated llms.txt prompts and tutorials.
  • Improved installation instructions and model definitions in examples.
  • Added proper docstrings to exposed methods and classes for better developer experience.

🐛 Bug Fixes

  • Model Warmup: Fixed an issue regarding the warmup sequence for model components.
  • Router API: Fixed SemanticRouter handling of action routes to align with the new Events/Actions API.

Full Changelog: 0.5.0...0.5.1

Assets 8
Loading