Things used in this project

Hardware components

Hiwonder MentorPi M1

×1

Software apps and online services

WonderPi

Story

Have you ever imagined owning a robot that not only follows commands but truly understands what you want to explore? Traditional robots might get you from point A to point B, but what if it could genuinely see the world around it and converse with you like a partner?

Meet MentorPi, an open-source robotic platform built on the Raspberry Pi 5 and ROS 2. It’s far more than just a SLAM navigation rover; it’s an intelligent agent deeply integrated with multimodal AI large models (language, vision, speech), merging precise low-level motion control, robust environmental perception, and high-level cognitive reasoning into a single, hands-on system.

From "Following Commands" to "Understanding Intent"

Imagine saying to it:

"Hey Mentor, first go to the zoo and see what animals are there; then head to the supermarket to check out what fruits are available; finally, take me to the soccer field for a game."

"Hey Mentor, first go to the zoo and see what animals are there; then head to the supermarket to check out what fruits are available; finally, take me to the soccer field for a game."

In traditional human-robot interaction, executing this seamlessly is nearly impossible—it contains three distinct layers of tasks:

• Semantic Location Navigation (Zoo, Supermarket, Soccer Field)
• Visual Cognitive Tasks upon Arrival (Identify animals, Identify fruits)
• Final Intent Understanding (Confirm the field is ready for play)

MentorPi accomplishes this coherently, thanks to the synergy between semantic understanding from large models and its SLAM navigation system.

How Does the Tech Stack Collaborate?

1. Task Comprehension & Planning

Voice commands are captured and converted to text.

A Language Large Model deconstructs the natural language instruction, extracting the three locations and their associated visual tasks to generate a structured mission queue.

2. Autonomous Navigation & Obstacle Avoidance

The SLAM system (using LiDAR and a prior map) handles point-to-point navigation.

Orchestrated by ROS 2, the robot plans optimal paths, moves reliably, avoids obstacles, and reaches each target area in sequence.

3. Visual-Semantic Understanding

Upon arrival, the Vision Large Model activates, scanning the scene via a 3D depth camera.

At the Zoo: It doesn’t just detect "animals" but provides a detailed description: "The scene includes models of a giraffe, kangaroo, tiger, etc."

At the Supermarket: It focuses on identifying fruits, reporting: "Various fruits are available, such as apples, bananas, grapes, and oranges. You can choose based on preference."

This represents an evolution from merely "seeing" to "comprehending" the scene’s semantic content.

4. Task Completion & Closure

Arriving at the soccer field, the robot confirms the user’s intent is satisfied, reporting "Arrived at the soccer field, ready to play!" and closing the task loop.

Why This is a Project Worth Building Yourself

MentorPi is not just a demo; it’s a fully open-source platform designed for learning and development:

Open Hardware: Based on Raspberry Pi 5 & ROS 2, it’s highly extensible and compatible with various sensors.

Flexible AI Integration: Supports either local lightweight models or cloud-based AI APIs (like GPT-4V), allowing you to balance performance and cost.

Modular Design: Clear separation between SLAM, voice interaction, visual recognition, and navigation modules makes debugging and customization easier.

Learning-Friendly: An ideal platform for advancing your skills in robotics, SLAM, 3D vision, human-robot interaction, and AI integration.

The Spark of Integration: Where Spatial Coordinates Meet Semantic Meaning

MentorPi’s breakthrough lies in its deep fusion of precise spatial positioning from SLAM ("where am I") with rich semantic understanding from AI models ("what is here, what is this place"). This transforms the robot from a simple tool executing "go to coordinates (x, y)" into a responsive "exploration partner" that interacts meaningfully with its environment.

What Can You Do With It?

• Explore cutting-edge research areas like Vision-and-Language Navigation (VLN).
• Develop more natural human-robot dialogue systems.
• Experiment with long-horizon task execution from complex instructions.
• Use it as a testing platform for robotics algorithms (path planning, dynamic obstacle avoidance, semantic mapping).
• Extend its capabilities—add a robotic arm to create an "see-understand-act" closed loop.

Conclusion

We believe the future of robotics lies not in faster movement or more precise grasping, but in how well robots understand our world and how naturally we can collaborate with them. MentorPi is our practical step in that direction, and we hope it becomes a starting point for more developers, students, and enthusiasts to enter the exciting field of Embodied AI.

Let’s turn robots from mere tools into curious extensions for exploring the world around us.

**Read more

Schematics

Credits