File size: 4,692 Bytes
473a751 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 | # Copyright (c) 2022-2025, The Isaac Lab Project Developers.
# All rights reserved.
# SPDX-License-Identifier: BSD-3-Clause
"""Per-task language prompts for VLA data collection / LeRobot conversion.
Keyed by the task slug produced by :func:`_task_slug` in both
``collect_vla_data.py`` and ``convert_to_lerobot.py``. Accessed via
:func:`resolve_prompt_from_task`, which normalizes any Gym id (e.g.
``ConveyorBeltPickAndPlace-Franka-DataCollection-v0``) into the same slug.
Keep the wording short, action-driven, and English-only; VLA tokenizers
handle these best. If a slug is missing, we fall back to
``LEGACY_DEFAULT_PROMPT`` so both scripts degrade gracefully.
"""
from __future__ import annotations
# Legacy default used as both the argparse default in collect_vla_data.py
# and the sentinel for "user did not customize the prompt" when deciding
# whether to auto-fill from the task lookup.
LEGACY_DEFAULT_PROMPT: str = "manipulate the object"
#
# Keys match the slug produced by ``_task_slug`` in ``collect_vla_data.py``
# and ``convert_to_lerobot.py`` (suffixes + robot tokens stripped, then joined
# by underscore). We also register the "spelled-out" variant (e.g.
# ``conveyor_belt_pick_and_place``) as a synonym for humans who call the resolver
# directly with task profile names.
_CONVEYOR_BELT_PICK_AND_PLACE = (
"pick up the cube from the moving conveyor and place it into the bin"
)
_WHACK_A_MOLE = "hit the mole that pops up on the board with the closed gripper"
_BALL_CATCHING = (
"hold the container with the gripper and catch the thrown ball in the container"
)
_ROLLING_BALL_INTERCEPTION = "catch the ball rolling down the ramp with the container"
_ROTATING_PEG_INSERTION = "insert the peg into the hole on the rotating disc"
_BALL_THROWING = "throw the ball into the bin"
TASK_PROMPTS: dict[str, str] = {
# slug as produced by `_task_slug(gym_id)`:
"conveyorbeltpickandplace": _CONVEYOR_BELT_PICK_AND_PLACE,
"whackamole": _WHACK_A_MOLE,
"ballcatching": _BALL_CATCHING,
"rollingballinterception": _ROLLING_BALL_INTERCEPTION,
"rotatingpeginsertion": _ROTATING_PEG_INSERTION,
"ballthrowing": _BALL_THROWING,
# human-readable profile names (also accepted):
"conveyor_belt_pick_and_place": _CONVEYOR_BELT_PICK_AND_PLACE,
"whack_a_mole": _WHACK_A_MOLE,
"ball_catching": _BALL_CATCHING,
"rolling_ball_interception": _ROLLING_BALL_INTERCEPTION,
"rotating_peg_insertion": _ROTATING_PEG_INSERTION,
"ball_throwing": _BALL_THROWING,
}
def _slugify(task_name: str) -> str:
"""Normalize a task/gym id into a lookup slug.
Mirrors ``_task_slug`` in ``collect_vla_data.py`` / ``convert_to_lerobot.py``:
strips known suffixes and robot tokens, then joins with underscores.
Also accepts already-slugified inputs like ``conveyor_belt_pick_and_place``.
"""
if not task_name:
return ""
normalized = task_name.lower().strip()
# Already a slug (no dashes) -> just return as-is (after normalizing underscores).
if "-" not in normalized and "_" in normalized:
return normalized
for suffix in ("-datacollection-v0", "-play-v0", "-ik-abs-v0", "-ik-rel-v0", "-v0"):
if normalized.endswith(suffix):
normalized = normalized[: -len(suffix)]
break
tokens = [
token for token in normalized.split("-")
if token and token not in {"franka", "panda", "ur5", "ur10", "xarm", "kinova"}
]
if not tokens:
return normalized.replace("-", "_")
return "_".join(tokens)
def resolve_prompt_from_task(
task_name: str | None,
default: str = LEGACY_DEFAULT_PROMPT,
) -> str:
"""Return the prompt registered for *task_name*, else *default*."""
if not task_name:
return default
slug = _slugify(task_name)
return TASK_PROMPTS.get(slug, default)
def resolve_prompt(
task_name: str | None,
user_prompt: str | None,
legacy_default: str = LEGACY_DEFAULT_PROMPT,
) -> str:
"""Decide which prompt to use.
Priority:
1. Non-empty *user_prompt* that differs from the legacy default.
2. Task-specific prompt from :data:`TASK_PROMPTS`.
3. *legacy_default*.
This means the user can still override via ``--prompt "my custom text"``,
while leaving ``--prompt`` untouched (or equal to the legacy default)
triggers per-task auto-fill.
"""
if user_prompt is not None and user_prompt != "" and user_prompt != legacy_default:
return user_prompt
return resolve_prompt_from_task(task_name, default=legacy_default)
|