Video generation

[spMohanty]

• Round-1 (extrinsic phase) : we will generate the videos from 5 randomly selected trials from the extrinsic phase.
• round-2 (intrinsic phase) : yet to be decided.
  • could be a timelapse
  • could be snippets of the activity from various phases of training
  • could be a realtime stream ? (this would be really cool !!)

[emilio-cartoni]

* Round-1 (extrinsic phase) : we will generate the videos from 5 randomly selected trials from the extrinsic phase.

* round-2 (intrinsic phase) : yet to be decided.

  * could be a timelapse
  * could be snippets of the activity from various phases of training
  * could be a realtime stream ? (this would be **really** cool !!)

Round 2: Timelapse (i.e. 1 frame every 1000) would not show much I think, you would only see arm and objects flashing around. There could be some changes, i.e. more objects moving towards the end and objects starting to be on the shelf if we are really lucky. Snippets: I would say get 1000 consecutive frames every 1M timesteps. This should show a more understandable video, with arms moving more randomly at the start and then more purposefully at the end. Realtime stream: Very cool! With the final evaluation taking up to 60h there would be a lot to stream :) I think something like a snippet video would still be needed to see it after (or we save the stream on youtube or somewhere else?)