---
license: apache-2.0
library_name: lerobot
pipeline_tag: robotics
tags:
  - robotics
  - vla
  - pi0.5
  - assembly
  - isaac-sim
base_model: DAVIAN-Robotics/pi05_droid_jointpos
datasets:
  - hud-evals/AssembleBench
---

# pi0.5 AssembleBench - CG-DAgger round 3 (final)

The final checkpoint from the writeup: the BC baseline plus three rounds of code-gated DAgger. This is what `examples/run_eval.py` loads by default.

pi0.5 finetuned for [AssembleBench](https://github.com/hud-evals/assemble-bench), a
14-task contact-rich robot assembly benchmark on the NIST ATB-1 taskboard: insert round
and square pegs, mesh gears, thread nuts. Built on Isaac Lab Arena with the DROID robot
platform (Franka Panda + Robotiq 2F-85).

Read the writeup: [Benchmarking Robot Models on Contact-Rich Assembly](https://www.hud.ai/blog/assemble-benchmark).

## Usage

```bash
git clone https://github.com/hud-evals/assemble-bench.git
cd assemble-bench
pip install -r examples/requirements.txt

# Serve the simulator (see the repo README for the Isaac Sim install)
OMNI_KIT_ACCEPT_EULA=YES python -m hud.environment.server env.py --port 8765

# Evaluate this checkpoint
python examples/run_eval.py --task peg_round_16mm --checkpoint hud-evals/pi05-AssembleBench-cgdagger-r3
```

The DROID input/output convention is not baked into the checkpoint. `examples/pi05_agent.py`
carries the adapter: front and wrist RGB resize-padded to 224, an 8-D state of 7 joint
positions plus gripper, and 8-D absolute joint-target actions at 15 Hz with a binarized
gripper.

## Training

Finetuned from [`DAVIAN-Robotics/pi05_droid_jointpos`](https://huggingface.co/DAVIAN-Robotics/pi05_droid_jointpos)
(full finetune, VLM included) on
[`hud-evals/AssembleBench`](https://huggingface.co/datasets/hud-evals/AssembleBench),
1355 synthetic demonstrations generated by scripted experts reading privileged simulator
state, with no teleoperation.

Code-gated DAgger (CG-DAgger): the policy drives, code detects failure and hands control to a scripted expert that demonstrates a recovery, and the policy is retrained on those intervention segments. It is HG-DAgger with the human replaced by a failure gate. Corrections were collected only on `peg_round_8mm`, so the other round-peg sizes measure whether the recovery transfers.

Three rounds were collected. Round 1 targeted grasp failure (grip lower and reattempt),
round 2 off-center peg placement (re-center before inserting), and round 3 insertion
failure from releasing the grasp mid-insertion (insert without ever opening the gripper).
Later rounds improved the looser clearances (12 mm and 16 mm); 4 mm stayed flat.

## Limitations

Task-specific: this is not a generalist checkpoint, and it is trained purely in simulation.
Success on these tasks is sparse - a handful of wins out of thirty on the harder families -
so the writeup reports dense reward alongside binary success. Square pegs and small gears
remain close to zero.

## Related checkpoints

| Checkpoint | What it is |
|---|---|
| [`pi05-AssembleBench-12k`](https://huggingface.co/hud-evals/pi05-AssembleBench-12k) | behavior-cloning baseline, all 14 tasks |
| [`pi05-AssembleBench-cgdagger-r3`](https://huggingface.co/hud-evals/pi05-AssembleBench-cgdagger-r3) | BC + 3 rounds of code-gated DAgger, the final checkpoint |