How to Score Higher on Arctivar V2
Less intervention can mean more valuable data. The score rewards two things: - Completing the rollout efficiently. - Keeping human intervention as brief as poss
Less intervention can mean more valuable data.
The score rewards two things:
- Completing the rollout efficiently.
- Keeping human intervention as brief as possible.
That means a higher-scoring contribution usually follows four steps:
- Let the policy work
- Intervene when necessary
- Correct only what is missing
- Hand control back when the policy recovers. If it doesn’t, take over and finish the task.
Now, let’s see how to put it into practice.
- Minimize takeover time: Let the robot run autonomously and intervene only when it is about to fail.
- Avoid full manual control: Focus on making precise corrections instead of completing the entire task yourself.
- Set up a clear camera view: Use an approximately 45-degree overhead angle and zoom in on the gripper and target object.
- Anticipate failures: Watch the robot’s trajectory carefully and take over only at the critical moment.
- Reset quickly: If an attempt clearly goes wrong, restart instead of wasting your takeover budget on a difficult recovery.
- Check the Goal Reference: Understand the expected result before starting to reduce unnecessary trial and error.
Watch the tutorial below for more practical tips.