GSoC 2026 - Interest in Project #21 - VLA Model Optimization - Aryan Jain IIT Madras

[aryannzzz]

Hi @nacui @baiheliu,

I'm Aryan Jain, IIT Madras, applying for GSoC 2026 Project #21
(Optimize a VLA model for embodied intelligence).

I have direct VLA experience, in my GRASP project I identified and fixed a gradient collapse bug caused by mean-pooling in VLA architectures, introducing attention-based pooling and reducing cross-instruction saliency by 85%. I also extended Grad-CAM to VLA policies and deployed on a real 6-DOF arm.

I couldn't find any open Good First Issues on the board to take. Could you suggest a prerequisite task I can work on, or would you accept a PR to the openvino_notebooks repo demonstrating a VLA model inference pipeline? I know I am pretty late for this but I never knew you worked on VLA arcitectures, and I will be glad if I could work on anything!

GitHub: https://github.com/aryannzzz

[baihe-liu]

Hi, thank you for your interest in this project. At the moment, I don't have any tasks to assign to you.
Although it is getting late, your proposal will still be considered if you submit it before the deadline.
Please refer to the existing discussion here for more detailed information.