molmoact-libero

Name: molmoact-libero
Author: graph-robots

26stars4forks

How to install molmoact-libero?

graph-robots/open-robot-skills/molmoact-libero

$npx -y skills add graph-robots/open-robot-skills --skill molmoact-libero

Installs into the current project.

Summary

Run the MolmoAct LIBERO checkpoint

Files · 1

View on GitHub

SKILL.md

1	# molmoact-libero
2
3	Closed-loop VLA-policy skill backed by one model checkpoint: AllenAI's
4	MolmoAct LIBERO checkpoint (`allenai/MolmoAct-7B-D-LIBERO-0812`). The skill
5	is the model — it owns its serving preset (`molmoact-libero`), so a policy
6	node names this skill, not a free-floating `policy_id`. The closed-loop
7	replan/execute/terminate body and the load-bearing LIBERO observation
8	encoding live in `gap.runtime.policy.run_policy_loop`; the websocket client
9	is resolved (and cached per preset) through the executor's `PolicyExecutor`.
10
11	This is the MolmoAct alternative to `pi05-libero` for the same task
12	family — the two are the policy A/B axis the benchmark ablates. Pick whichever
13	the task / experiment calls for; their capability envelope is the same.
14
15	## Capability
16
17	- Embodiment: Franka Panda (LIBERO/robosuite), OSC_POSE delta action
18	space `[Δx, Δy, Δz, Δrx, Δry, Δrz, gripper]`. No embodiment translation
19	happens in the loop — the checkpoint's native action space is forwarded to
20	`sim.apply_policy_action`.
21	- Tasks: the LIBERO pick-and-place distribution — pick a tabletop rigid
22	object, optionally place/drop it in a container. Works best steered:
23	perceive the target and hover the end-effector above it (preserving the
24	current rotation) before handing over, so the policy starts in-distribution.
25	- Not for: deformables / cloth folding, articulated objects, non-LIBERO
26	embodiments, or tasks the checkpoint never saw. If the task is outside this
27	envelope, pick a different skill or report a missing capability — do not
28	delegate it here and hope.
29
30	## Serving
31
32	The bundle ships its own `server.py` and declares MolmoAct-flavored openpi
33	as a git dep in its own `pyproject.toml`, so the bundle is self-contained:
34	no `$GAP_OPENPI_DIR` clone, no shared venv. First-run setup is
35	`gap skills install molmoact-libero`, which `uv sync`s the bundle's `.venv/`
36	with vLLM + MolmoAct deps. The launcher then spawns the server via
37	`uv run --project policies/molmoact-libero -- python server.py ...` (so the
38	bundle's own venv activates automatically) and downloads the checkpoint from
39	`hf://allenai/MolmoAct-7B-D-LIBERO-0812` on first run.
40
41	The bundle's `server.py` is a placeholder that documents how to wire
42	up a vLLM-style server speaking the openpi websocket protocol; replace it
43	with your real serving script (e.g., from an internal MolmoAct fork) before
44	running the bundle for the first time. Run it yourself with
45	`gap policy serve molmoact-libero`. A `policies:` config entry named
46	`molmoact-libero` overrides the recipe (e.g. an external `url:`).
47
48	## Termination & exits
49
50	The loop exits on whichever fires first — a commanded gripper
51	open→close→open cycle (`gripper_cycle`, the per-item terminator for
52	clean-all loops; set `gripper_cycle_termination: true`), a non-empty
53	`termination_prompt` answered yes by the VLM (`completed_by_vlm`

People who liked this also viewed

6 picks

Type

ai-video-generationinference.sh Agent skills for using our API to give your agents access to hundreds of apps and other agentsSkillsJul 2026292k642
imagegen-frontend-webElite frontend image-direction skill for generating premium, conversion-aware website design references. CRITICAL OUTPUT RULE — generate ONE separate horizontal image FOR EVERY section. A landing pageSkillsJul 2026165k66k
hyperframes-mediaWrite HTML. Render video. Built for agents.SkillsJul 2026145k37k
media-useAgent Media OS, the single skill for every media need in a HyperFrames project. Resolve BGM, SFX, image, icon, brand logo, voice, color grade, or LUT into a frozen local file or paste-ready block + leSkillsJul 2026144k37k
general-videoAuthor or edit a custom HyperFrames composition when no specialized workflow fits, or when BRIEF.md sets flow: companion. Use for longer or multi-scene pieces, brand and sizzle reels, montages, staticSkillsJul 2026124k37k
product-launch-videoTurn a product or marketing URL, pasted script, or brief into a product launch / promo video — SaaS promos, feature reveals, product demos, app and company launches. Use when the user wants to market,SkillsJul 2026121k37k

molmoact-libero

26stars4forks

View on GitHub

How to install molmoact-libero?

graph-robots/open-robot-skills/molmoact-libero

$npx -y skills add graph-robots/open-robot-skills --skill molmoact-libero

Installs into the current project.

Summary

Run the MolmoAct LIBERO checkpoint

Files · 1

View on GitHub

SKILL.md

1	# molmoact-libero
2
3	Closed-loop VLA-policy skill backed by one model checkpoint: AllenAI's
4	MolmoAct LIBERO checkpoint (`allenai/MolmoAct-7B-D-LIBERO-0812`). The skill
5	is the model — it owns its serving preset (`molmoact-libero`), so a policy
6	node names this skill, not a free-floating `policy_id`. The closed-loop
7	replan/execute/terminate body and the load-bearing LIBERO observation
8	encoding live in `gap.runtime.policy.run_policy_loop`; the websocket client
9	is resolved (and cached per preset) through the executor's `PolicyExecutor`.
10
11	This is the MolmoAct alternative to `pi05-libero` for the same task
12	family — the two are the policy A/B axis the benchmark ablates. Pick whichever
13	the task / experiment calls for; their capability envelope is the same.
14
15	## Capability
16
17	- Embodiment: Franka Panda (LIBERO/robosuite), OSC_POSE delta action
18	space `[Δx, Δy, Δz, Δrx, Δry, Δrz, gripper]`. No embodiment translation
19	happens in the loop — the checkpoint's native action space is forwarded to
20	`sim.apply_policy_action`.
21	- Tasks: the LIBERO pick-and-place distribution — pick a tabletop rigid
22	object, optionally place/drop it in a container. Works best steered:
23	perceive the target and hover the end-effector above it (preserving the
24	current rotation) before handing over, so the policy starts in-distribution.
25	- Not for: deformables / cloth folding, articulated objects, non-LIBERO
26	embodiments, or tasks the checkpoint never saw. If the task is outside this
27	envelope, pick a different skill or report a missing capability — do not
28	delegate it here and hope.
29
30	## Serving
31
32	The bundle ships its own `server.py` and declares MolmoAct-flavored openpi
33	as a git dep in its own `pyproject.toml`, so the bundle is self-contained:
34	no `$GAP_OPENPI_DIR` clone, no shared venv. First-run setup is
35	`gap skills install molmoact-libero`, which `uv sync`s the bundle's `.venv/`
36	with vLLM + MolmoAct deps. The launcher then spawns the server via
37	`uv run --project policies/molmoact-libero -- python server.py ...` (so the
38	bundle's own venv activates automatically) and downloads the checkpoint from
39	`hf://allenai/MolmoAct-7B-D-LIBERO-0812` on first run.
40
41	The bundle's `server.py` is a placeholder that documents how to wire
42	up a vLLM-style server speaking the openpi websocket protocol; replace it
43	with your real serving script (e.g., from an internal MolmoAct fork) before
44	running the bundle for the first time. Run it yourself with
45	`gap policy serve molmoact-libero`. A `policies:` config entry named
46	`molmoact-libero` overrides the recipe (e.g. an external `url:`).
47
48	## Termination & exits
49
50	The loop exits on whichever fires first — a commanded gripper
51	open→close→open cycle (`gripper_cycle`, the per-item terminator for
52	clean-all loops; set `gripper_cycle_termination: true`), a non-empty
53	`termination_prompt` answered yes by the VLM (`completed_by_vlm`

People who liked this also viewed

6 picks

Type