Research Scientist Intern, Wearable Sensing & Human Motion Analysis

Meta; Redmond (WA); Presencial; USD 55 000 - 83 000; Tempo integral

Resumo da oferta

Meta's Reality Labs Research Surreal Vision internship focuses on developing a learned motion analysis model for swing sports, using egocentric data from Aria Gen 2 and Neural Band wearables. You will work on timing segmentation, pose estimation, and swing-quality scoring in a lab setting.

The twelve to twenty-four week program emphasizes hands-on data collection, model development, and evaluation against a baseline, with opportunities for publication or internal tech reports.

Qualificações

• PhD in CS/EE/Robotics or related field in progress or completed.
• Strong Python and PyTorch experience.
• Experience with sequence modeling and real sensor data.
• Willing to conduct hands-on experiments with participants.
• Authorized to work in US at time of hire.

Responsabilidades

• Operate egocentric data collections with lab simulator and wearables.
• Design and train models over multimodal time-series data.
• Tackle swing-phase segmentation and swing-quality scoring.
• Build evaluation harness comparing to threshold baselines.
• Define capture/labeling protocols for small, noisy datasets.
• Integrate winning model into prototype and demo it end-to-end.
• Write results; potential for publication or internal report.

Conhecimentos

PyTorch

Python

Time-series modeling

Sensor data

Human testing

Work authorization

IMU data

Publications

Formação académica

PhD in CS/EE/Robotics

Ferramentas

Wearable sensors

Motion capture hardware

Aria Gen 2

Descrição da oferta de emprego

About

The Meta Reality Labs Research Team brings together a world-class team of researchers, developers, and engineers to create the future of AR and VR. The Surreal Vision group at RL Research is seeking exceptional Research Scientists to research and help build the egocentric machine perception functionalities and digital twins of 3D scenes that will underpin future VR and contextual-AI-enabled AR devices.This internship focuses on developing a learned model for motion analysis for swing sports as an alternative to heuristic, threshold-based methods. Research goals include: improving analysis accuracy, generalization across players and equipment, and generating interpretable feedback outputs. The intern will contribute to egocentric data collection using the team's simulator research setup, in addition to model development.Our internships are twelve (12) to twenty-four (24) weeks long and we have various start dates throughout the year. Some projects may require a minimum of 24 consecutive weeks.

Responsibilities

• Operate egocentric data collections on our state-of-the-art in-lab simulator - instrument participants with Aria Gen 2 and Neural Band wearables, run capture sessions, and use the simulator's ground-truth as supervision and validation signal for your models.
• Design and train models over multimodal egocentric time-series: Aria Gen 2 (SLAM/IMU, eye gaze, RGB) fused with Neural Band surface EMG and wrist IMU.
• Tackle the core modeling problems: temporal swing phase segmentation (address → takeaway → impact → follow-through), kinematic/intent estimation from a head-mounted and wrist-worn viewpoint, and swing-quality scoring.
• Build an evaluation harness that benchmarks the model against the existing threshold baseline on held-out players - accuracy, phase-boundary error, and cross-subject generalization.
• Define the capture and labeling protocol, and handle the realities of small, noisy, personalized datasets (self-supervised pretraining, augmentation, subject-adaptive fine-tuning).
• Integrate the winning model back into our prototype and demo it end to end.
• Write up results; strong work is a candidate for publication or an internal tech report.

Minimum Qualifications

• Currently has, or is in the process of obtaining a PhD degree in the domain of Computer Science, Electrical Engineering, Robotics, Biomechanics, or a related field
• Strong Python and hands-on experience training models in PyTorch
• Working knowledge of sequence/time-series modeling (temporal CNNs, transformers, state-space models, HMM/CTC-style segmentation)
• Experience working with real sensor data, not just curated benchmarks
• Comfortable running hands-on hardware experiments with human participants in a lab setting
• Must obtain work authorization in the country of employment at the time of hire, and maintain ongoing work authorization during employment Experience designing and executing your own data collections with wearable or motion-capture hardware
• Research experience with IMU-based human activity recognition, pose/motion estimation, or wearable biosignals
• Experience with surface EMG: filtering, rectification, envelope extraction, electrode-shift and cross-session variability
• Familiarity with egocentric perception (Project Aria, Ego4D/Ego-Exo4D, Aria Digital Twin) or multimodal sensor fusion
• Background in sports biomechanics or motion analysis
• self-supervised or few-shot learning for personalization
• First-author publications at CVPR/ICCV/ECCV/NeurIPS/ICLR, or in biomechanics and wearables venues
• Plays swing sports - genuinely helpful for labeling and sanity-checking feedback
• Intent to return to a degree-program after the completion of the internship