Roadmap

​

Legged Robots

Quadruped and humanoid robot manifests are the next major robot type expansion. This includes channel definitions for multi-leg gait controllers, balance feedback loops, and new simulation containers with appropriate physics. The channel interface already supports arbitrary joint counts, so the core work is in manifest definitions, simulation containers, and gait-specific safety limits.

​

Real Hardware Validation

roz currently validates safety exclusively in simulation. Moving to real hardware requires additional safety layers:

• Hardware-in-the-loop (HIL) testing with physical e-stop circuits
• Gradual authority transfer (simulation-verified controllers run at reduced power before full deployment)
• Sensor fault detection (distinguish “sensor reports zero” from “sensor is disconnected”)

This is the most safety-critical item on the roadmap.

​

gz-transport-rs Proto Coverage

The gz-transport-rs crate currently covers 83 of approximately 180+ Gazebo transport protobuf definitions. Expanding coverage enables richer simulation feedback — camera images, point clouds, contact forces, and other sensor modalities that the agent can use for spatial reasoning. Contributions here are straightforward: each proto is a standalone addition with codegen and a thin Rust wrapper.

​

Skills Marketplace

A registry for community-contributed agent skills — reusable tool bundles that extend what the agent can do. Examples:

• Welding path planners
• Pick-and-place strategies for specific gripper types
• Inspection routines for common industrial parts

Skills are implemented as TypedToolExecutor implementations and can be loaded at runtime.

​

Visual Spatial Reasoning

Camera stream integration for vision-based tasks. The agent would receive image frames from simulation or hardware cameras and use multimodal LLM capabilities for:

• Object detection and pose estimation
• Visual servoing (closed-loop control guided by camera feedback)
• Scene understanding for task planning

This depends on LLM provider support for real-time image streams alongside text context.

​

ROS 2 Native Integration

Today, roz communicates with ROS 2 through a Zenoh bridge running inside Docker simulation containers. A native ROS 2 integration would allow direct topic subscription and service calls without the bridge layer. This removes the Docker requirement for teams that already have a ROS 2 workspace, reduces latency, and simplifies the deployment stack for production hardware.