Deployment¶

This guide covers deploying and evaluating LBM robotics policies using the inference utilities provided in VLA Foundry.

Overview¶

The vla_foundry/inference/scripts/ directory provides lightweight scripts for serving trained LBM robotics policies over gRPC. Each script is ready to run with uv so that you can reuse the repository's managed environment.

Script	Description
`launch_wave_policy.sh`	Launches a dummy gRPC policy server that waves the robot end-effectors in a simple sinusoidal pattern. Useful for verifying the gRPC stack.
`launch_inference_policy.sh`	Downloads a DiffusionPolicy checkpoint from S3 (if needed) and launches a gRPC policy server that uses it.

Prerequisites¶

Before running the deployment scripts, ensure the following:

Complete the project setup in the repository root (see main README for uv sync --frozen instructions).
Install the inference group: uv sync --group inference.
Provide any required credentials (e.g., AWS, W&B, Hugging Face tokens) for accessing checkpoints or datasets referenced in your configuration.
Run scripts from the repository root.

Wave Policy Demo¶

The wave policy is a deterministic scripted policy that validates the gRPC stack.

Launch the wave policy server¶

bash vla_foundry/inference/scripts/launch_wave_policy.sh

Key behavior:

Starts a gRPC server that streams sinusoidal joint poses to connected clients until interrupted.

Inference Policy Demo¶

launch_inference_policy.sh handles both the checkpoint download and the server launch in a single step.

Launch the inference policy server¶

From the repository root:

bash vla_foundry/inference/scripts/launch_inference_policy.sh <experiment_name>

For example:

bash vla_foundry/inference/scripts/launch_inference_policy.sh \
  2025_11_05-21_34_11-model_diffusion_policy-lr_5e-05-bsz_1024

To pin a specific checkpoint number (the highest available is selected by default):

bash vla_foundry/inference/scripts/launch_inference_policy.sh <experiment_name> 5

The S3 root and download destination can be overridden via the VLA_MODELS_S3 and VLA_INFERENCE_DEST_DIR environment variables. See vla_foundry/inference/scripts/README.md for the full list of options.

Connecting a client¶

These scripts only start the policy server. To exercise it end-to-end, point your gRPC simulation client (for example, lbm_eval) at the running server.