Visual World Models Engineer for Robot Learning

Ashby, Inc.; Cambridge (MA); In loco; USD 120.000 - 180.000; Tempo pieno

Descrizione del lavoro

Ashby, Inc. is seeking a hands-on researcher or engineer to develop visual world models for robots in Cambridge, MA. You will focus on learning from images, video and interaction data to predict scene dynamics and guide robot behaviour with action-conditioned models.

You will work closely with robotics and controls engineers to connect representations to planning and policy learning, conducting rigorous experiments and addressing latency, computation and generalization challenges on real robots.

Competenze

• Hands-on experience in world models and video representation learning.
• Strong Python and PyTorch skills with ownership of training/evaluation.
• Solid foundation in computer vision and deep learning for temporal/sequential data.
• Experience designing rigorous experiments with baselines and clear analyses.
• Interest in translating research into capabilities on real robots with engineers.

Mansioni

• Build and evaluate visual world models capturing scene dynamics and robot actions.
• Develop models from video and robot data, incl. self-supervised learning and action-conditioned prediction.
• Link learned representations to planning and robot policies with cross-functional teams.
• Own experiments from data collection to model training, evaluation, and validation on physical systems.
• Design evaluations for predictive accuracy, robustness, and usefulness for downstream tasks.
• Investigate failures, improve data/models, and address latency and compute constraints.

Conoscenze

World models

Video representation learning

Predictive modelling

PyTorch

Python

Computer vision

Deep learning

Temporal modelling

Experimentation

Robotics experience

Strumenti

ROS

NumPy

Descrizione del lavoro

Ashby, Inc. is seeking a hands-on researcher or engineer to develop visual world models for robots in Cambridge, MA. You will focus on learning from images, video and interaction data to predict scene dynamics and guide robot behaviour with action-conditioned models.

You will work closely with robotics and controls engineers to connect representations to planning and policy learning, conducting rigorous experiments and addressing latency, computation and generalization challenges on real robots.