[GSoC]Project #10 discussion: Agentic Dora — MVP scope and simulation workflow

[DGHX12345]

Hello dora-rs team,

I am very interested in Project #10: Agentic Dora — Agent-Driven Robot Control for GSoC.

From my understanding, the goal of this project is to connect an agent framework with Dora so that users can control a simulated robot through high-level commands, without manually wiring low-level pipelines.

I would like to discuss a possible MVP direction, since this project seems to have a large design space.

My current idea is to keep the first version focused on a simple but complete loop:

1. a simulated robot environment (e.g., PyBullet)
2. a Dora pipeline exposing sensors and actuators as nodes
3. an agent bridge node that converts high-level commands into control actions
4. a minimal set of robot actions (e.g., move, stop, read sensor)
5. a reproducible end-to-end demo

To keep the project realistic, I am currently thinking about:

• using a mock agent implementation first (rule-based or simple planner), before integrating more complex frameworks
• limiting robot capabilities to a small set of actions
• focusing on a single simulation scenario rather than a general system

I would like to ask a few questions to better understand expectations:

1. For the MVP, what is more important:
   • integration with Dora dataflows, or
   • the sophistication of the agent side?
2. Would you recommend starting with a mock agent first, then extending to real frameworks later?
3. How complex should the simulation environment be for the first version?
4. Should the demo focus more on agent interaction, or on Dora integration and dataflow design?

My goal is to make the first version small, reliable, and demonstrable, rather than trying to cover too many features.

Thank you!

[KOKOSde]

Your MVP scope looks right, but I would start with a mock or rule based agent bridge instead of a real LLM. That lets you prove the Dora message flow, simulator loop, and action schema first, which is the real risk.

[DGHX12345]

Thanks — that makes a lot of sense.

I agree that the main risk is validating the Dora message flow, simulator loop, and action schema. I’ll start with a mock or rule-based agent bridge and focus on building a reliable end-to-end pipeline first.

I’ll refine the proposal based on this direction and submit a more detailed plan soon.