Block Stacking Order Agnostic¶
Agent runs
Each mode of this task runs once per run. Est. cost is tokens at list price (data/prices.yml), never a bill; see the RoboLab runs for every task of a run.
Codex + GPT-6 Luna, reasoning xhigh (ChatGPT login)default run
| Trial | Started | Agent time | Model requests | Tokens in / out | Est. cost | Grade |
|---|---|---|---|---|---|---|
| U✓ 10m | 09-27 02:53 | 10m | 31 | 1.3M / 9k | $0.024 | deterministic 1, grader_error 0, instance_exact 1, instance_ok 1, n_actions 489, video_rendered 1 |
| L✓ 9m | 09-29 11:31 | 9m | 23 | 657k / 17k | $0.020 | deterministic 1, grader_error 0, live_success 1, n_actions 303, replay_success 1, video_rendered 1 |
Task instruction (upstream)
Stack the blocks into a tower
No oracle video published upstream — this is the scene it starts from.
What RoboLab states about this task
| Success predicate | stacked |
| Predicate arguments | objects: ['red_block', 'blue_block', 'green_block', 'yellow_block'] · order: None |
| Scored subtasks | 3 |
| Subtask predicates | Subtask stacked |
| Objects | red_block blue_block green_block yellow_block |
| Upstream attributes | stacking |
| Upstream difficulty | simple |
| Episode budget | 90 s |
| Other wordings | vague — Build a tower specific — Pick up the colored blocks and stack them vertically on top of each other, forming one tower |
| Environment class | BlockStackingOrderAgnosticTask |
| Upstream name | BlockStackingOrderAgnosticTask |
| Defined in | robolab/tasks/benchmark/block_stacking_order_agnostic_task.py |
From https://github.com/NVlabs/RoboLab @ v0.3.1.
Tags¶
Task DomainManipulation
Skill primitives in the demo: stacking
Why this task is interesting¶
Not yet written.
Capability notes¶
Not yet written.
Oracle demo review¶
Not yet reviewed.