---
tags:
  - robotics
  - vla
  - machine-unlearning
  - pi0.5
  - libero
license: cc-by-4.0
---

# Behavior Uncloning — Entangled-Object PO Step 80

**Experiment**: Entangled-object unlearning (Pivot A)
**Task**: Forget "milk" (T5) while retaining visually near-identical "orange juice" (T0) on LIBERO-Object
**Method**: PO (Action Redirection (PO))
**Base model**: pi0.5 (3.35B params, flow matching)

## Results

| Metric | Value |
|--------|-------|
| **HM** | **1.00** |
| Forget SR (milk) | 0% |
| Retain SR | 100.0% |
| OJ SR (visual neighbor) | 100% |

## Context

This checkpoint is from the **entangled-object forgetting** experiment, which tests
whether unlearning can distinguish between visually near-identical objects (milk and
orange juice are both rectangular cartons of similar size).

Unlike the [butter experiment](https://huggingface.co/haohw/behavior-uncloning-object-po-step80)
where the forget target was geometrically unique, this experiment deliberately uses
the hardest visual discrimination case in the dataset.

See the [full report](https://github.com/haohww/behavior-uncloning/blob/main/docs/experiments/vla/entangled_object_report.md) for details.

## Training Config

- Suite: LIBERO-Object, Forget: T5 (milk, dataset index 5)
- Steps: 80, BS=32, LR=1e-5
- Hardware: 8xA100-80GB (FSDP)
- Framework: [openpi](https://github.com/Physical-Intelligence/openpi)