EVA-Client: A Unified Data Collection, Inference, and Deployment Framework for Embodied Policies on Real Robots

2026-07-02

AI Summary

This work addresses the fragmented deployment, data collection, and evaluation pipelines that hinder efficient real-world policy iteration in robotics. We propose a unified framework featuring a decoupled component architecture that orthogonally combines robot backends, inference policies, and communication middleware. This design supports diverse inference strategies—including synchronous/asynchronous execution, ACT temporal integration, and real-time chunked inference—and unifies Debug, Collect, and Eval workflows. By implementing “evaluation-as-collection,” the framework automatically records complete trajectories and system logs, generating training-ready datasets and visual comparison reports. The resulting traceable and reusable data loop significantly accelerates real-robot policy development and iteration.

Abstract

We present EVA-Client, an open-source framework for deployment, data collection, and evaluation of trained manipulation policies on real robots. Sitting between a policy server and the physical hardware, EVA-Client unifies the real-robot stages of the policy iteration loop within a single codebase. It makes three contributions. First, a component-decoupled architecture in which robot backends, inference strategies, and transport middlewares form an orthogonal grid: adding a robot or a strategy touches only its own layer. Second, inspectable execution through Debug, Collect, and Eval workflows, with modes ranging from open-loop simulation to continuous real-time control. Third, every evaluation run doubles as a data collection, recording full rollouts in training-ready format alongside exhaustive logs and a side-by-side comparison viewer, so each evaluation feeds the next round of training rather than ending as an unrecorded impression. EVA-Client further consolidates major real-time inference strategies, synchronous and asynchronous execution, ACT-style temporal ensembling, Real-Time Chunking, and a naive-async ablation baseline, behind a single configuration surface.

Problem

• embodied policies
• real robots
• data collection
• policy deployment
• evaluation framework

Innovation

• component-decoupled architecture
• inspectable execution
• training-ready data collection
• real-time inference strategies
• policy deployment