Multi-Agent Reinforcement Learning Research Scientist

Pheratech Systems, Inc.; Los Angeles (CA); On-site; USD 140,000 - 210,000; Full time

Job summary

Pheratech Systems, Inc. in Los Angeles, CA is seeking a Research Scientist to advance model-based reinforcement learning for coordinated robotic systems. You will work with the research lead on world models, consensus, and networked dynamics, with ownership over research questions and direction.

The role emphasizes learning objective design, experimentation, and collaboration with simulation staff to connect scientific advances to robotic coordination across timescales.

Qualifications

• PhD in machine learning, robotics, CS, applied mathematics, physics, control, or related.
• Post-PhD research experience of at least one year in academia or industry.
• Strong publication record with original scientific contributions.
• Deep expertise in model-based RL and learning dynamics.
• Expertise in multi-agent representations and decentralized execution.
• World modeling, state estimation, and planning under uncertainty.
• Strong statistical validation, experimental design, and ablations.
• Excellent technical communication and collaborative proposal writing.
• Awareness of experimental best practices and rigorous record-keeping.

Responsibilities

• Originate research questions and develop novel algorithms and representations.
• Develop world models and model-based RL for interacting, heterogeneous agents.
• Investigate communication, partial observability, changing teams, and coupled dynamics.
• Develop uncertainty-aware exploration and planning with limited data.
• Explore hierarchical and temporal representations across timescales.
• Adapt small VLMs and connect multimodal grounding to world models and planning.
• Design rigorous experiments to test hypotheses and compare explanations.
• Collaborate with researchers and simulation engineers to apply advances to robotics.
• Lead projects through implementation, validation, and scientific publication.
• Attend conferences/workshops and assist with proposals.

Skills

Model-based RL

Multi-agent representations

World modeling

State estimation

Planning under uncertainty

Latent representations

Small VLMs

Multimodal grounding

Hierarchical representations

Temporal representations

Python

PyTorch

JAX

Scientific communication

Proposal writing

Experimental design

Uncertainty analysis

Education

PhD in ML/CS/related

Tools

PyTorch

JAX

Job description

Research Scientist — Model-Based Reinforcement Learning and Multi-Agent Systems

Pheratech · Los Angeles, California

Pheratech is seeking a Research Scientist to develop new methods for learning, prediction, and decision-making in coordinated robotic systems.

This scientist will work closely with the existing research lead on problems connecting learned world models with consensus, coordination, and networked dynamics. The role offers substantial ownership of research questions, methodological development, and scientific direction, with dedicated simulation staff supporting experimentation.

The central specialization is model-based reinforcement learning, with research spanning multi-agent representations, world modeling, state estimation, and planning under uncertainty. The work emphasizes latent representations that capture decision-relevant structure and support coordinated behavior across agents and timescales. It also includes compact vision-language models and their integration with grounded state, action, and task representations.

Research responsibilities

• Originate research questions and develop novel algorithms, representations, learning objectives, or theoretical insights.
• Develop world models and model-based RL methods for interacting, heterogeneous agents.
• Investigate how communication, partial observability, changing team composition, and coupled dynamics affect prediction and control.
• Develop uncertainty-aware exploration and planning methods that use limited interaction data effectively.
• Investigate hierarchical and temporal representations, including skill discovery, memory, and planning across multiple timescales.
• Adapt small VLMs and connect multimodal representations to world models, planners, and robotic behavior.
• Design statistically rigorous experiments that distinguish competing explanations, test assumptions, and establish the limits of a method.
• Collaborate with researchers and simulation engineers to connect scientific advances with robotic coordination.
• Lead research projects through implementation, validation, technical writing, and publication where appropriate.
• Attend conferences and workshops and assist with writing proposals.

Required qualifications

• A completed PhD in machine learning, robotics, computer science, applied mathematics, physics, control, or a related discipline.
• At least one year of post-PhD research experience in academia or industry.
• A strong publication record demonstrating substantive, original scientific contributions, including research initiated and intellectually led by the candidate.
• Deep expertise in model-based RL, including learned dynamics, imagined policy learning, planning, model exploitation, and the relationship between prediction quality and control performance.
• Expertise in multi-agent representations, including relational structure, symmetry, local information, communication, heterogeneous teams, and decentralized execution.
• Strong understanding of world modeling and state estimation, including action-conditioned prediction, latent states, belief representations, stochastic dynamics, and partial observability.
• Working knowledge of dynamical systems and networks, including stability, feedback, delays, graph structure, and coupled dynamics.
• Strong statistical analysis and validation skills, including experimental design, uncertainty estimation, effect sizes, ablations, sensitivity analysis, and generalization testing.
• Sound scientific judgment: clear hypotheses, defensible assumptions, informative baselines, and willingness to revise conclusions in response to evidence.
• Knowledge of uncertainty and exploration, including calibration, epistemic and aleatoric uncertainty, information‑seeking behavior, and planning under model uncertainty.
• Experience with small VLMs and multimodal grounding, including adaptation, resource‑aware evaluation, and connecting semantic representations to state, action, and planning.
• Knowledge of hierarchical and temporal representations, including skill discovery, temporal abstraction, memory, and planning across multiple timescales.
• Proficiency in Python and a modern ML framework such as JAX or PyTorch, with demonstrated ability to implement, debug, and reproduce research methods.
• Clear technical communication and constructive engagement in scientific discussion and disagreement.
• Extensive experience writing collaborative proposals for academic or government funding.
• Awareness of experimental best practices and record‑keeping methods.

The ideal candidate will also have proficiency in a low‑level programming language. A demonstrated track record of securing academic funding as PI, Investigator, or Key Personnel is optional but highly valued, as is a history of valuable contributions to high‑impact open‑source research software.