Coding week12&13 8/12-8/25

In recent weeks, we have focused on enhancing model evaluation, deployment workflows, video presentation quality, and looking for next research objectives. The August 14 and 21 meetings highlighted our progress and set new goals, the meeting details are summarized in Google Doc.

Model Evaluation Challenges

In the August 14 meeting, regarding the research idea, we discussed the model’s performance in handling varied scenarios, especially focusing on collision avoidance mechanisms. Achieving high predictive accuracy remains a significant challenge, as it is essential for enhancing the model’s reliability across various simulated environments. Currently, our primary focus is on improving model accuracy through expanding the diversity of training user instruction data and scenario configurations.

Model Deployment

To address GitHub storage limitations and Hugging Face compatibility, we proposed streamlined workflows for easier updates. Key improvements include integrating OpenAI API support to simplify model loading. Next steps are to complete testing and documentation for the OpenAI API. See PR here

Action Items

Based on current works, the following tasks were outlined as priorities:

• Immediate Fixes: Address issues in an open Pull Request on the CARLA branch, resolve Streamlit errors, integrate OpenAI API, and fix model loading challenges.
• Research and Documentation: Conduct a literature review and document findings for research continuity.
• Social Media Engagement: After reviewing the 1-minute demo video, it was agreed that the captions required adjustment for clarity, and background music will be added to increase appeal. Once these enhancements are complete, I will draft a LinkedIn post summarizing project achievements and outcomes to accompany the video. Draft a LinkedIn post to accompany the 1-minute demo to show recent outcomes.