
The podcast explores the complexities of designing approachable software systems for robotics, particularly balancing capability with ease of use. Sebastian Castro shares his experiences and mistakes in robotic software engineering, advocating for systems that are simple to understand and supported by documentation and community. He contrasts engineered and learned systems, highlighting the interpretability challenges of machine learning. Castro uses examples from his work, including a task planning architecture and motion planning frameworks, to illustrate the importance of logging, visualization, and testing. He emphasizes not reinventing the wheel, getting user buy-in before optimizing, and prioritizing the ability to troubleshoot, concluding that understanding the human element is key to solving technical challenges.
Sign in to continue reading, translating and more.
Continue