Robot Data & Evaluation Tooling Intern

# Robot Data & Evaluation Tooling Intern Palo Alto, CA Internship

On-site $20-$75/hour based on experience

Posted July 30, 2026

Apply for this role About the role## Overview Real Hand  is an AI robotics company building dexterous robotic hands and the intelligence that drives them. The goal of the company is to give machines human-level dexterity — hardware that survives real-world contact, and policies learned from how humans actually use their hands. Our hands are deployed on humanoid and mobile platforms across research, automation, and human-robot collaboration.

Real Hand is headquartered in Palo Alto, CA.We are looking for a Robot Data & Evaluation Tooling Intern to own the infrastructure behind our egocentric and teleoperation datasets, and the evaluation harness that tells us whether a new policy is actually better. Dexterous manipulation is a data problem as much as an algorithms problem, and the quality of this pipeline sets the ceiling on everything the rest of the team can do.

If you think the eval harness is the most important artifact in an ML organization, this role is for you.

This is an on-site internship with a direct path to a full-time role.

Salary range: $20-$75/hour based on experience.

What you will do## Responsibilities
* Build scalable pipelines for ingesting and processing multi-modal robot and human data: egocentric video, wrist cameras, depth, joint states, tactile and force signals, teleoperation streams, and audio — with time synchronization that actually holds
* Design annotation strategies and tooling for hand-object interaction: hand pose and contact labeling, object segmentation and tracking, grasp type and phase segmentation, task and subtask boundaries, and success/failure labeling
* Build model-assisted and semi-automated labeling loops (SAM-family segmentation, hand-pose estimators, trackers, active learning) and the QA workflow that keeps their output honest
* Own dataset management: schema design, versioning, lineage, deduplication, augmentation, and tooling that lets a researcher slice the corpus by task, platform, operator, or failure mode in seconds
* Build the evaluation framework — standardized benchmark suites in physical simulation and on real hardware, automated regression reporting, per-behavior success metrics, and dashboards the team actually reads
* Generate and validate synthetic data in simulation, and quantify how its distribution differs from real-world data
* Implement reliability tooling: data drift detection, sensor failure detection, uncertainty estimation, and anomaly flagging in the collection pipeline
* Curate and analyze public egocentric datasets and assess their transfer value to our platforms
* Partner with the loco-manipulation, manipulation, and perception teams to turn their pain points into tooling

What you bring## Requirements
* Strong Python and solid software engineering practice: testing, code review, reproducibility
* Experience building data or ML pipelines at non-trivial scale, and evaluation frameworks for ML systems
* Working knowledge of computer vision — detection, segmentation, tracking, depth estimation, or 3D scene understanding — with PyTorch, Tensor Flow, or JAX
* Familiarity with egocentric video and hand-object interaction annotation, including the specific difficulties of first-person data rather than video data in general
* Experience generating or working with data from a physics simulator, and an understanding of how synthetic and real data differ in distribution
* Ability to design experiments, analyze failures, and iterate quickly in real-world robotic systems
* Thrive in fast-paced, ambiguous environments where solutions require exploration and ownership

Nice to have## Bonus qualifications
* Currently enrolled in an MS or PhD program in Robotics, Computer Science, EE, or a related field
* Familiarity with robot data formats and ecosystems: ROS bags, LeRobot, RLDS, Open X-Embodiment
* Familiarity with egocentric datasets such as Ego4D, EgoExo4D, EPIC-KITCHENS, HOI4D, or DexYCB
* Uncertainty modeling, calibration, and failure detection
* Data visualization and dashboard building (Rerun, Foxglove, Streamlit, or similar)
* C++ familiarity for high-throughput data tooling
* Passion for building robots that do real physical work outside the lab