Founding Research Scientist, Robot Learning

Our first research frontier is self-preservation: the base case of physical self-replication. We are building a new class of machines called insectoids that can survive, coordinate, and recover without humans. We believe scalable machine labor requires more than single-agent task generality or machines shaped in our image.

About the role

GRAM is building reusable embodied intelligence that transfers across embodiments, tasks, environments, tools, and team configurations. The work must remain grounded in data, compute, runtime constraints, and repeatable physical evaluation.

You will be the accountable owner of GRAM’s robot-learning research agenda. You will set technical direction and evaluation standards, make architecture, data, and compute tradeoffs directly with the founders, and shape the hiring standard and mentor the team as the program grows. You will develop general representations, models, policies, training methods, and evaluations for individual and coordinated physical behavior. Success means measured transfer to physical systems, not breadth asserted from a benchmark.

What you will do

• Develop, pretrain, and adapt robot foundation models that acquire reusable physical capabilities from heterogeneous experience.
• Set the research roadmap, technical standards, and experimental decision process for GRAM’s robot-learning program.
• Design representations, objectives, architectures, and adaptation methods that transfer across robot morphology, sensor configuration, task, environment, tool, and machine variation.
• Build training curricula from heterogeneous physical and simulated experience, with strict dataset and checkpoint lineage.
• Scale experiments in PyTorch or JAX while separating gains from model architecture, objective, data composition, compute, initialization, and evaluation leakage.
• Define frozen evaluations and falsifiable capability claims for task transfer, environmental robustness, embodiment transfer, data efficiency, latency, and recovery.
• Deploy selected models through C++ robotics runtimes, then use physical failures and offline-to-online discrepancies to determine the next research question.
• Help recruit, evaluate, and mentor the researchers and engineers who extend the program.

• PhD in machine learning, robotics, computer science, applied mathematics, or a related field, or an equivalent record of original research demonstrated by publications, research systems, or deployed capabilities that can be examined during the hiring process.
• Led a consequential technical direction in robot learning, reinforcement learning, imitation learning, or embodied foundation models and can show how your decisions changed the resulting system or research program.
• Strong Python and PyTorch or JAX skills, plus working C++ ability for model integration, profiling, and real-time inference.
• Trained an embodied model or policy using a versioned dataset and evaluated it on held-out tasks, environments, embodiments, tools, or agent configurations defined before model selection.
• Deployed a learned model on a physical robot, autonomous vehicle, or other closed-loop physical system; can present measured performance, the validation design, and a failure that changed the research direction.

Preferred experience

• Vision-language-action models, transformer or diffusion policies, offline reinforcement learning, imitation learning, or self-supervised representation learning.
• Technical agenda-setting, research hiring, mentoring, or establishing evaluation standards for an early research program.
• Distributed training, active data collection, sim-to-real transfer, closed-loop fleet learning, or large-scale evaluation systems.

The annual base salary range for this San Francisco position is $225,000â??$300,000. An offer within this range will reflect the position’s approved scope and the candidate’s demonstrated role-relevant skills and experience.

Working at GRAM

This role is based on-site in San Francisco with direct access to physical robots. The research loop extends from data and training through deployment, measurement, and failure analysis on physical machines.

GRAM expects deep trust and ownership from its people, and we begin by extending the same to candidates. We treat your information, prior work, and conversations with discretion.