Inspiration
In an age where technology is increasingly compartmentalized with multiple components working in tandem, we wanted to return to the roots. The beauty of end-to-end solutions lies in their simplicity and elegance. Inspired by comma.ai's vision of pushing the boundaries of what's possible, we envisioned an indoor navigation solution that's truly end-to-end. Enter Luigi, our end-to-end trained robot. The church of end2end is not just a project; it's a testament to the power of holistic solutions.
What it does
Luigi, the comma body, is trained to navigate indoor courses seamlessly. Using a dataset comprising commands like "turn left", "turn right", and "straight", Luigi processes real-time inputs and linearly interpolates between the model outputs for these classifications. This allows Luigi to steer itself and navigate any indoor path presented to it, making decisions that are akin to an intuitive understanding of its environment.
How we built it
We started by collecting a diverse set of data within indoor environments, encompassing a myriad of scenarios Luigi might encounter. This dataset was then used to train our model using deep learning techniques, ensuring that it could generalize well across different indoor settings. The linear interpolation between model outputs served as our control system, bridging the gap between discrete commands and continuous navigation requirements.
Challenges we ran into
Training a model that can interpolate between classifications with high accuracy was a challenge. Ensuring that Luigi responded smoothly, especially during transitions (e.g., moving from a "turn left" scenario to a "straight" path) required iterative refinement. We also had to account for various indoor obstacles and ensure that Luigi's decisions were both safe and efficient.
Accomplishments that we're proud of
Achieving a truly end-to-end indoor navigation system was our primary goal, and we're thrilled to have realized it. Watching Luigi navigate complex indoor terrains, making split-second decisions, and adapting to unexpected changes in the environment brought immense satisfaction. The seamless integration of the training data, the model, and the control system is something we're particularly proud of.
What we learned
Beyond the technical aspects, we learned the importance of perseverance. True innovation often requires challenging established norms and pushing boundaries. We also gained insights into the nuances of indoor navigation, understanding that it's not just about moving from point A to B, but about doing so with 0 feature engineering.
What's next for church of end2end
The church of end2end has just begun its journey, and the future looks promising.
A Note on Creation
In the spirit of true end-to-end solutions, even this post was generated holistically with the power of GPT-4. No manual edits, no intermediary steps, just a direct conversation with the model. The church of end2end truly believes in and practices what it preaches.
Built With
- e2e
- end2end
Log in or sign up for Devpost to join the conversation.