`stage0_smoke@v1` — Stage-0 tri-embodiment smoke rig

Three heterogeneous ManiSkill agents — a Panda arm, a Fetch mobile manipulator, and an Allegro hand — share one scene and step at different control rates. The task proves the multi-agent rig (env construction, per-agent action routing, deterministic seeding) runs end to end. It carries no cooperation claim; it exists so harness failures are caught before any benchmark task runs.

Tier 0 — rig diagnostics · Status: DIAGNOSTIC · Suite: CHAMBER-Bench v1.0 (adr/ADR-027-chamber-bench-v1-protocol.md)

Spaces

Action space. Dict over three uids (panda_wristcam, fetch, allegro_hand_right) stepping at heterogeneous control rates (20/10/50 Hz).
Observation. Per-agent ManiSkill state observations.
Simulator backend. maniskill3 · Agents. 3

Stress channel

None — no force coupling channel.

Axis validity

Per-cell status under the validity matrix (adr/ADR-027-chamber-bench-v1-protocol.md):

AS	OM	CR	CM	PF	SA
`untested`	`untested`	`untested`	`untested`	`untested`	`untested`

Industrial analogue

The commissioning dry-run of a mixed-vendor robot work cell: before production starts, integrators step every controller through its motion envelope to prove the cell's plumbing — power, signals, cycle timing — rather than its productivity.

Evidence

How to run one episode

import chamber.tasks

env = chamber.tasks.make("stage0_smoke")
obs, info = env.reset(seed=0)
# Requires a SAPIEN-capable host (Tier-2).

Notes

Tier-0 rig diagnostic: proves the multi-agent harness (env construction, per-agent action routing, deterministic seeding) runs end to end; carries no cooperation claim (ADR-027 §Tier ladder). No dedicated smoke archive exists under spikes/results/ — the executable evidence is the Stage-0 acceptance suite (make smoke) plus ADR-001.

stage0_smoke@v1 — Stage-0 tri-embodiment smoke rig