File size: 3,043 Bytes
8aa2acf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# evaluation_pi0_pi05

Evaluation code for **π0** and **π0.5** (OpenPI) on ManiSkill robotic manipulation tasks.

This repository contains two branches, one per model family:

| Branch | Model | Source |
|--------|-------|--------|
| `pi0.5` | π0.5 (OpenPI inference) | [yqi19/Openpi_inference](https://github.com/yqi19/Openpi_inference) |
| `pi0`   | π0 (OpenPI finetune)    | [yqi19/pi0_finetune](https://huggingface.co/yqi19/pi0_finetune)     |

---

## Supported Evaluation Tasks

Both branches contain inference/evaluation code for the following ManiSkill experiments:

| Directory | Description |
|-----------|-------------|
| `examples/maniskill/` | General random-attribute ManiSkill evaluation |
| `examples/maniskill_conflict/` | OOD conflict experiment evaluation (factor dominance) |
| `examples/maniskill_attribute/` | Single-attribute pairwise factor evaluations |
| `examples/maniskill_full_factor/` | Full-factor (all_factor) evaluation |
| `examples/maniskill_verb_color_object/` | Verb × color × object 3-factor evaluation |
| `examples/maniskill_verb_size/` | Verb × size 2-factor evaluation |

---

## Branch: `pi0.5`

π0.5 evaluation code using the OpenPI inference server.

### Quickstart

```bash
# 1. Start the policy server (pi0.5)
uv run scripts/serve_policy.py --env MANISKILL \
    policy:checkpoint \
    --policy.config pi05_maniskill \
    --policy.dir <checkpoint_dir>

# 2. Run conflict experiment evaluation
bash examples/maniskill_conflict/run_ood_experiment_inference.sh \
    <num_episodes> <results_file>

# 3. Run verb×color×object evaluation
bash examples/maniskill_verb_color_object/run_verb_color_object_inference.sh \
    <num_episodes> <results_file>

# 4. Run full-factor evaluation
bash examples/maniskill_full_factor/run_full_factor_inference.sh

# 5. Run all factor evaluations (multi-seed)
bash run_multi_seed_evals.sh
bash scripts/eval_openpi_port8000.sh   # port 8000 worker
bash scripts/eval_openpi_port8001.sh   # port 8001 worker
```

### VLM Evaluation
```bash
python scripts/vlm_eval_all_conflicts.py
```

---

## Branch: `pi0`

π0 evaluation code.

### Quickstart

```bash
# 1. Start the policy server (pi0)
uv run scripts/serve_policy.py --env MANISKILL \
    policy:checkpoint \
    --policy.config pi0_maniskill \
    --policy.dir <checkpoint_dir>

# 2. Run conflict experiment evaluation
bash run_pi0_conflict_eval.sh

# 3. Run all-factor evaluation scripts
bash eval_all_9_ckpts.sh       # all 9 checkpoints
bash eval_all_ckpts.sh         # all available checkpoints
bash run_all_3seeds.sh         # 3-seed evaluation

# 4. Run full-factor evaluation
bash examples/maniskill_full_factor/run_full_factor_inference.sh
```

### VLM Evaluation
```bash
python vlm_eval_pi0_conflicts.py
```

---

## Notes

- Checkpoint paths must be specified at runtime; no model weights are included in this repo.
- Both branches share the same core `src/openpi/` library and `examples/` structure.
- Evaluation results are written to a path you specify; no results data is committed here.