Plan and execute one-step AI2-THOR agent navigation via MCP tools.
npx skills add https://github.com/MiaoDX/roboclaws --skill ai2thor-navigator
Please help me install this Agent Skill. Skill: ai2thor-navigator Source: https://github.com/MiaoDX/roboclaws/tree/main/skills/ai2thor-navigator Command: npx skills add https://github.com/MiaoDX/roboclaws --skill ai2thor-navigator
Navigate a simulated AI2-THOR agent one step at a time using MCP tools to perform controlled, minimal-move navigation in indoor scenes.
Start by issuing an observe call to initialize the MCP session, then run a loop of move and observe to advance toward your goal.
High-intent search queries and answers about installing and using this skill.
To navigate an AI2-THOR agent step by step using MCP tools, initialize an observe call, then run a loop of move and observe actions to advance toward your goal. This provides per-step feedback and frame data to validate progress in indoor scenes.
Single-step navigation in AI2-THOR simulation environments is a deterministic process issuing one movement action per decision cycle. It applies to single-agent indoor tasks, requiring environment observation, path planning, and per-step movement commands with feedback.
Yes, AI2-THOR navigation with MCP requires the roboclaws__scene_objects, roboclaws__observe, roboclaws__move, roboclaws__goto, and roboclaws__done tools. These facilitate environment observation, path planning, and per-step movement commands following the MCP protocol with explicit yaw conventions.
Yes, you can use MCP tools for obstacle avoidance in AI2-THOR indoor scenes. By issuing one movement action per decision cycle and utilizing per-step feedback and frame data, you can validate progress and adjust your strategy to avoid obstacles.
The limitation of single-action navigation in AI2-THOR is that it only supports single-agent indoor navigation tasks. It requires per-step movement commands and explicit yaw conventions, making it suited for controlled, minimal-move testing and demos rather than continuous multi-agent scenarios.
To plan and execute a path to a target in an AI2-THOR scene, start with an observe call to initialize the MCP session, then use move and goto commands to advance toward your goal, finishing with a done call to complete the navigation.
taomiaoScan Alfworld observations to catalog visible objects and receptacles.
code-review-benchmarkCreate a first-person camera with quaternion orientation and keyboard/mouse input for SDL3 GPU scenes.
ddalcuAutomate web browser interactions for navigation, observation, and action execution.
taomiaoPlan next navigation actions toward target receptacles in household environments.
