# How I built my own intelligent Robot Arm from Scratch - Iulia Feroli - NDC Oslo 2026

## Executive summary

The speaker details the journey of building an intelligent robot arm from basic electronics to an advanced, AI-powered prototype. The talk emphasizes the field of Physical AI, which integrates LLMs and AI techniques into physical robotics and embedded systems. The process highlights iterative development, starting with low-cost components (like cardboard and basic servos) and progressing through advanced stages involving 3D printing, soldering, and integrating Python/FastAPI for movement control and data recording.

## Key takeaways

- Physical AI Definition: Physical AI is a new generation of robotics applications that bring AI models and sensors into the physical world, allowing autonomous systems to generate movement instructions without requiring painstaking, complex code or physics-engine implementations.
- Iterative Hardware Development: The speaker advocates for building prototypes from scratch (e.g., using cardboard, then plastic, then 3D-printed parts) to learn fundamental concepts and avoid expensive, pre-built kits.
- AI Integration via Function Calls: Intelligence can be added by defining specific functions (e.g., 'say hi', 'grab') and having an LLM agent decide which functions to call, rather than requiring full model retraining for every task.
- Data Collection for Training: The process of recording movement episodes (teleoperation) is crucial for training models, allowing the robot to eventually perform tasks autonomously, even if the initial control is human-guided.

## Technical details

- Microcontrollers and Computing Boards: Options for the 'brain' include microcontrollers like Arduino, microprocessors like Raspberry Pi (or the smaller Raspberry Pi Zero), and specialized boards like Nvidia boards (for CPU/GPU capabilities). The choice depends on budget and required processing power.
- Actuators and Control: Joint movement is typically achieved using servo motors. The speaker notes the cost difference between basic $1 servos and advanced actuators (up to $40-$60 per joint) that provide state reporting (position feedback).
- Prototyping and Assembly: The speaker transitioned from initial cardboard prototypes to plastic, and finally utilized 3D printing (using open-source models) to create a durable, custom-fit body for the robot arm, significantly reducing costs compared to commercial kits.
- Software and Control Logic: The system is controlled using Python, with functions created and exposed via FastAPI. Movement is defined by telling each servo motor its target position, which can be recorded and played back.
- Advanced Integration: The final stage involves connecting the system to an LLM loop, allowing the agent to decide which physical functions (e.g., 'say hi') to execute based on a conversation, demonstrating physical embodiment of AI.

## Practical implications

- Begin robotics projects with low-cost, modular components (e.g., basic servos and microcontrollers) to learn core principles.
- Utilize 3D printing and open-source designs to build custom, cost-effective hardware bodies for robotic systems.
- Implement AI functionality by defining discrete, callable functions (APIs) for the robot, rather than attempting to train a single, massive model initially.
- Understand that data collection (recording movement episodes) is the foundational step for training autonomous robotic models.

## Topics

Robotics, Physical AI, Embedded Systems, Machine Learning, LLMs, Mechatronics, 3D Printing, Hugging Face, ROS (Robotic Operating System), Save Studio

Source: https://www.youtube.com/watch?v=iFqYI18Rpn8