Getting out of the way: my robotics crash course (thisismypersonalblog.com)

In a thought-provoking exploration of robotics, a seasoned software developer chronicles his recent venture into LLM-assisted robotic programming over a nine-month period. The author, initially skeptical about integrating robotics into his life, decided to purchase a preassembled 3D printed robot arm to engage his children in hands-on technology. His journey, fueled by tools like Codex and later Fable, showcases the rapid progress he made in developing software that not only controls the robot but also learns through interaction. Despite initial difficulties, such as power limitations and miscalculations during motion, the LLM eventually enabled the robot to perform tasks like picking and placing colored blocks based on visual feedback—a notable achievement. This experience highlights the potential of LLMs in practical robotics applications, bridging the gap between theoretical programming and tangible results. While many within the AI/ML community concentrate on abstract models, the author’s project emphasizes the significance of real-world testing and user-friendly interfaces for engaging both technical and non-technical audiences. By harnessing these advancements, he aims to refine the system further, exploring the transition from LLMs to Variable-length Architectures (VLA) for enhanced efficiency and cost-effectiveness. The implications of his work suggest a promising future where AI not only assists in programming but also substantially elevates the capabilities of robotic systems.