`mpe_cooperative_push@v1` — MPE cooperative push (CPU diagnostic)

Two point agents on a unit-square plane cover two landmarks under a shared cooperative-coverage reward, using continuous 2-D velocity actions. The env is a deterministic, dependency-free equivalent of the classic MPE simple-spread task and runs on any CPU. It is the substrate for the ego-AHT empirical-guarantee experiment and for CI smoke coverage of the training loop.

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

Spaces

Action space. Dict over two uids (ego, partner); continuous 2-D velocity actions clipped to [-1, 1] per axis.
Observation. Per-agent numpy state vectors (agent + landmark positions) under the shared cooperative-coverage reward.
Simulator backend. pure_python · Agents. 2

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

Deliberately thin by design: it abstracts two mobile units jointly spreading over drop-off points on a floor plan, and exists as a CPU-friendly diagnostic of the learning harness rather than a model of any production operation.

Evidence

How to run one episode

import chamber.tasks

env = chamber.tasks.make("mpe_cooperative_push")
obs, info = env.reset(seed=0)
# Pure Python — runs on any host.

Notes

CPU rig / CI smoke substrate for the ego-AHT empirical-guarantee experiment (make empirical-guarantee); carries no cooperation claim (ADR-027 §Tier ladder). No run archive exists under spikes/results/ — the executable empirical-guarantee suite is the evidence.

mpe_cooperative_push@v1 — MPE cooperative push (CPU diagnostic)