Model Card for TBD-VLA

TBD-VLA (Temporal Block Diffusion Vision Language Action Model) is a Vision-Language-Action policy based on Block Discrete Denoising Diffusion. It uses a Qwen3-VL vision-language backbone and predicts robot action chunks through temporal-level block diffusion.

The model was introduced in the paper TBD-VLA: Temporal Block Diffusion Vision Language Action Model by Sung-Wook Lee, Xuhui Kang, and Yen-Ling Kuo.

• Project Webpage: https://tbd-vla.github.io/
• Code: LeRobot fork

This policy has been trained and pushed to the Hub using a LeRobot fork. See the full documentation here.

Results

LIBERO

Success rates (%) across the four LIBERO task suites.

LIBERO-Plus

Zero-shot success rates (%) across the seven LIBERO-Plus perturbation factors.

Inference Time

Action chunk generation latency measured on a single NVIDIA RTX A40 GPU.

How to Get Started with the Model

Installation

git clone https://github.com/TBD-VLA/lerobot.git
cd lerobot
uv python install 3.12
uv venv --python 3.12
source .venv/bin/activate
uv pip install -e ".[libero]"
uv pip install -U transformers
uv pip install -U accelerate

Train from scratch

python src/lerobot/scripts/lerobot_train.py \
  --policy.type=tbdvla \
  --output_dir=/$OUTPUT_DIR \
  --dataset.repo_id=sean1295/libero_all \
  --job_name=tbdvla_experiment \
  --steps=150000 \
  --batch_size=4 \
  --save_freq=20000 \
  --log_freq=1000 \
  --policy.device=cuda \
  --policy.n_bins=512 \
  --policy.block_temporal_size=4 \
  --policy.n_diffusion_steps=2 \
  --policy.gripper_dims=[-1] \
  --policy.chunk_size=16 \
  --policy.n_action_steps=16 \
  --policy.gradient_checkpointing=true \
  --policy.push_to_hub=false \
  --wandb.enable=false

Writes checkpoints to the configured output directory.

Multi-GPU training

accelerate launch --multi_gpu --num_processes=4 \
  src/lerobot/scripts/lerobot_train.py \
  --wandb.enable=false \
  --num_workers=4 \
  --policy.type=tbdvla \
  --policy.push_to_hub=false \
  --dataset.repo_id=sean1295/libero_all \
  --steps=150000 \
  --save_freq=20000 \
  --log_freq=1000 \
  --batch_size=16 \
  --policy.device=cuda \
  --policy.n_bins=512 \
  --policy.block_temporal_size=4 \
  --policy.n_diffusion_steps=2 \
  --policy.gripper_dims=[-1] \
  --policy.chunk_size=16 \
  --policy.n_action_steps=16

Evaluate the policy/run inference

uv run python src/lerobot/scripts/lerobot_eval.py \
  --policy.path=$CKPT_DIR \
  --env.type=libero \
  --env.task=libero_10 \
  --eval.n_episodes=50 \
  --eval.batch_size=1 \
  --eval.use_async_envs=false \
  --policy.device=cuda \
  --policy.n_action_steps=12 \
  --policy.n_diffusion_steps=2 \
  --policy.compile_model=true

Use --policy.path to point to a local or Hub checkpoint.

Real-Time Chunking (RTC)

TBD-VLA supports Real-Time Chunking (RTC) for latency-aware action generation. RTC allows the next action chunk to be generated while the current chunk is being executed and uses temporal in-painting to maintain consistency with previously generated actions.

In the LeRobot implementation, RTC latency compensation is controlled by --policy.latency_timestep. The value specifies the number of action timesteps corresponding to inference latency; the default value 0 disables latency compensation.

For example, to evaluate with a latency of 4 action timesteps:

uv run python src/lerobot/scripts/lerobot_eval.py \
  --policy.path=$CKPT_DIR \
  --env.type=libero \
  --env.task=libero_10_with_obs_delay_t4 \
  --eval.n_episodes=50 \
  --eval.batch_size=1 \
  --eval.use_async_envs=false \
  --policy.device=cuda \
  --policy.n_action_steps=12 \
  --policy.n_diffusion_steps=2 \
  --policy.latency_timestep=4

Set --policy.latency_timestep according to the measured inference latency and the control frequency of the environment or robot. Use 0 for standard chunked inference without RTC latency compensation.

Model Details

• Model type: Vision-Language-Action policy
• Architecture: Block Diffusion VLA
• VLM backbone: Qwen/Qwen3-VL-2B-Instruct
• License: apache-2.0

Architecture

TBD-VLA contains the following main components:

Files

Key Parameters

Model Architecture

Diffusion / Block Denoising

Training Hyperparameters

Inference Hyperparameters

BibTex

@article{lee2026tbdvlatemporalblockdiffusion,
      title={TBD-VLA: Temporal Block Diffusion Vision Language Action Model},
      author={Lee, Sung-Wook and Kang, Xuhui and Kuo, Yen-Ling},
      journal={arXiv preprint},
      year={2026},
      url={https://arxiv.org/abs/2606.07895}
}