Hi everyone,
I’ve been experimenting with some reinforcement Learning models focused on navigation and pathfinding within complex 3D environments, specifically using the Roblox engine as a testing ground because of its physics. To handle the agent’s interactions and automate certain repetitive tasks for data collection, I’ve been using deltaexector to run the custom Lua scripts that interface between my model and the game environment.
I’m running into a bit of a synchronization issue. When I’m logging metrics to my W&B dashboard, there seems to be a significant lag between the script execution in deltaexector and when the telemetry data actually hits my project logs. I’m trying to determine if the overhead from the executor is causing my training loops to de-sync or if it’s a bottleneck in how I’m handling the API requests within the scripts.
Has anyone here used similar third-party script tools for gathering training data in sandboxed environments? I’m specifically worried about whether the way deltaexector handles memory injection might be interfering with the local Python listener I have set up for my W&B runs. If anyone has tips on optimizing the bridge between mobile-based executors and experiment tracking, or if there’s a better way to ensure high-fidelity logging while the executor is active, I’d love to hear your thoughts!