Swarmbotics AI is hiring: Swarm Engineer - Multi-Agent Task Planning in Phoenix
Swarmbotics AI, Phoenix, AZ, United States
Company background
Swarmbotics AI is a low-cost, swarm robotics company for industry and defense. We see a world of ubiquitous low-cost robots transforming almost all aspects of society, but we see an urgent need in the defense industry. We focus on building swarms of robots that incorporate a low-cost BOM, an autonomous stack optimized for off the shelf components, and a global planner that enables swarm capabilities for groups of robots to accomplish sophisticated tasks.
Our first product is a defense application building Unmanned Ground Vehicles (UGVs), collectively termed - Attritable, Networked, Tactical Swarm (ANTS). Each UGV in ANTS is an independently-tasked, attritable robot designed for on-demand and autonomous mobility. When operating as a swarm, ANTS is capable of executing more advanced and coordinated, high-level capabilities across a battlespace. ANTS will help solve some of the DoD’s biggest problems that will save lives and increase defense capabilities.
Stephen Houghton and Drew Watson are the Founders and have decades of experience in self-driving cars and trucks, humanoids, and UAVs with experience from NASA, JPL, Cruise, Embark, McKinsey, Amazon, and the CIA.
Job description
Swarmbotics AI is seeking a Machine Learning Engineer to design, develop, and deploy a multi-modal action model that enables each UGV to select and execute coordinated swarm macro-actions in real time. This role sits at the intersection of machine learning and multi-agent decision making: you will build learned models that reason over multi-modal inputs to perform tactical macro-actions. This is not a perception role. The core focus is on the decision-making and action-selection layers — training models that translate situational awareness into intelligent swarm behavior. You will work closely with company leadership and cross-functional teams to align capabilities with the Swarmbotics AI product roadmap.
What You'll Do
Architect, train, and iterate on multi-modal action models that select swarm-level tactical macro-actions from rich contextual inputs
Design model architectures that fuse heterogeneous inputs — local perception, swarm state, mission objectives — into a unified decision representation
Develop and apply online and offline reinforcement learning approaches, including transformer-based sequence modeling, to learn swarm coordination policies
Optimize models to run real-time on edge devices through quantization, distillation, and efficient architecture design
Build and maintain the full pipeline from data collection and curation through training, evaluation, and field deployment
Integrate the action model into the broader autonomy stack alongside navigation, planning, and swarm coordination subsystems
Deploy and validate trained models on physical UGV swarms in field environments
Write robust Python and C++ code
Required qualifications
Strong mathematical foundation in neural networks, transformers, reinforcement learning, and statistics
Proficiency in Python and C++
Experience with PyTorch or TensorFlow
Experience training and deploying models that produce actions or macro-actions (e.g., online or offline reinforcement learning, planning-as-inference, VLA models, or similar) — not solely classification or perception
Familiarity with multi-agent coordination concepts: task allocation, distributed decision-making, or swarm behaviors
Experience optimizing and deploying ML models on resource-constrained or edge hardware
Preferred qualifications
Hands-on experience with policy gradient methods such as PPO
Experience with multi-agent task planning algorithms (task allocation, scheduling, auction-based methods)
Familiarity with ONNX, TensorRT, and edge deployment toolchains
Prior robotics experience, autonomous driving background, or work with unmanned systems
Experience with simulation environments and synthetic data generation for training multi-agent policies
Experience owning an entire data-to-production model pipeline
Academic publications in related fields (e.g., NeurIPS, AAAI, IROS, ICRA, JAIR)
Experience with a CatBs framework is preferred but not required
The preceding description is not designed to be a complete list of all duties and responsibilities required for the position. Swarmbotics is an equal‑opportunity employer. All qualified applicants will be treated with respect and receive equal consideration for employment without regard to race, color, caste, creed, religion, sex, gender identity, sexual orientation, national origin, ancestry, disability, uniform service, Veteran status, age, or any other protected characteristic per federal, state, or local law.
#J-18808-Ljbffr
In Summary: Swarmbotics AI is seeking a Machine Learning Engineer to design, develop, and deploy a multi-modal action model that enables each UGV to select and execute coordinated swarm macro-actions in real time . This role sits at the intersection of machine learning and multi-agent decision making . The core focus is on the decision-making and action-selection layers .
En Español: La empresa Swarmbotics AI es una compañía de robótica en enjambre de bajo costo para la industria y la defensa. Vemos un mundo ubicuo de robots de bajo coste que transforman casi todos los aspectos de la sociedad, pero vemos una necesidad urgente en la industria defensiva. Nos centramos en construir enjambres de robots que incorporan un BOM de bajo precio, una pila autónoma optimizada para componentes fuera del estante y un planificador global que permite capacidades de enjambre para grupos de robots para realizar tareas sofisticadas. Nuestro primer producto es una aplicación de defensa construyendo vehículos no tripulados terrestres (UGVs), colectivamente denominados - Attritable, Network Founded, Tactical Swarm (ANTS). Cada UGV en ANTS es un robot de acción ajetreable independiente diseñado para manejar a demanda y movilidad humana. Este papel se encuentra en la intersección del aprendizaje automático y la toma de decisiones multi-agentes: construirás modelos aprendidos que razonan sobre entradas multimodales para realizar macroacciones tácticas. Trabajará en estrecha colaboración con el liderazgo de la empresa y los equipos interfuncionales para alinear las capacidades con la hoja de ruta del producto Swarmbotics AI. What You'll Do Arquitecto, entrenamiento e iteración sobre modelos de acción multimodales que seleccionan macroacciones tácticas a nivel de enjambre desde fuertes entradas contextuales Diseñar arquitecturas de modelo que fusionan insumos heterogéneos percepción local, estado de enjambre, objetivos de misión en una representación unificada de decisiones Desarrollar y aplicar enfoques de aprendizaje de refuerzo online y offline, incluyendo modelado de secuencias basado en transformadores, aprender políticas de coordinación conjunta Optimizar modelos para ejecutar dispositivos en tiempo real a través de cuantización, destilación y eficiencia Conceptos de diseño de maquinaria de construcción y desarrollo de procesos más eficaces A partir de la recopilación de datos por medio de la tensión conjunta Tensión de trabajo mediante capacitación, evaluación y análisis de tareas de ejemplares Los propietarios de un proyecto de trabajo en línea desarrollan y aplican estrategias de desarrollo automático de sistemas de gestión de recursos humanos (PLAV) Optometría de proyectos o métodos de formación autónomos como la generación de conocimientos básicos y implementación de software, planificación de programas de programación de trabajo, planeación multitécnicas y configuración de computadoras y redes de Python, Experiencias avanzada o aplicaciones técnicas de aprendizajes, etc. Swarmbotics es un empleador de igualdad de oportunidades. Todos los solicitantes calificados serán tratados con respeto y recibirán la misma consideración para el empleo sin importar raza, color, casta, credo, religión, sexo, identidad de género, orientación sexual, origen nacional, ascendencia, discapacidad, servicio uniforme, estatus de veterano, edad o cualquier otra característica protegida por ley federal, estatal u local. #J-18808-Ljbffr
Swarmbotics AI is a low-cost, swarm robotics company for industry and defense. We see a world of ubiquitous low-cost robots transforming almost all aspects of society, but we see an urgent need in the defense industry. We focus on building swarms of robots that incorporate a low-cost BOM, an autonomous stack optimized for off the shelf components, and a global planner that enables swarm capabilities for groups of robots to accomplish sophisticated tasks.
Our first product is a defense application building Unmanned Ground Vehicles (UGVs), collectively termed - Attritable, Networked, Tactical Swarm (ANTS). Each UGV in ANTS is an independently-tasked, attritable robot designed for on-demand and autonomous mobility. When operating as a swarm, ANTS is capable of executing more advanced and coordinated, high-level capabilities across a battlespace. ANTS will help solve some of the DoD’s biggest problems that will save lives and increase defense capabilities.
Stephen Houghton and Drew Watson are the Founders and have decades of experience in self-driving cars and trucks, humanoids, and UAVs with experience from NASA, JPL, Cruise, Embark, McKinsey, Amazon, and the CIA.
Job description
Swarmbotics AI is seeking a Machine Learning Engineer to design, develop, and deploy a multi-modal action model that enables each UGV to select and execute coordinated swarm macro-actions in real time. This role sits at the intersection of machine learning and multi-agent decision making: you will build learned models that reason over multi-modal inputs to perform tactical macro-actions. This is not a perception role. The core focus is on the decision-making and action-selection layers — training models that translate situational awareness into intelligent swarm behavior. You will work closely with company leadership and cross-functional teams to align capabilities with the Swarmbotics AI product roadmap.
What You'll Do
Architect, train, and iterate on multi-modal action models that select swarm-level tactical macro-actions from rich contextual inputs
Design model architectures that fuse heterogeneous inputs — local perception, swarm state, mission objectives — into a unified decision representation
Develop and apply online and offline reinforcement learning approaches, including transformer-based sequence modeling, to learn swarm coordination policies
Optimize models to run real-time on edge devices through quantization, distillation, and efficient architecture design
Build and maintain the full pipeline from data collection and curation through training, evaluation, and field deployment
Integrate the action model into the broader autonomy stack alongside navigation, planning, and swarm coordination subsystems
Deploy and validate trained models on physical UGV swarms in field environments
Write robust Python and C++ code
Required qualifications
Strong mathematical foundation in neural networks, transformers, reinforcement learning, and statistics
Proficiency in Python and C++
Experience with PyTorch or TensorFlow
Experience training and deploying models that produce actions or macro-actions (e.g., online or offline reinforcement learning, planning-as-inference, VLA models, or similar) — not solely classification or perception
Familiarity with multi-agent coordination concepts: task allocation, distributed decision-making, or swarm behaviors
Experience optimizing and deploying ML models on resource-constrained or edge hardware
Preferred qualifications
Hands-on experience with policy gradient methods such as PPO
Experience with multi-agent task planning algorithms (task allocation, scheduling, auction-based methods)
Familiarity with ONNX, TensorRT, and edge deployment toolchains
Prior robotics experience, autonomous driving background, or work with unmanned systems
Experience with simulation environments and synthetic data generation for training multi-agent policies
Experience owning an entire data-to-production model pipeline
Academic publications in related fields (e.g., NeurIPS, AAAI, IROS, ICRA, JAIR)
Experience with a CatBs framework is preferred but not required
The preceding description is not designed to be a complete list of all duties and responsibilities required for the position. Swarmbotics is an equal‑opportunity employer. All qualified applicants will be treated with respect and receive equal consideration for employment without regard to race, color, caste, creed, religion, sex, gender identity, sexual orientation, national origin, ancestry, disability, uniform service, Veteran status, age, or any other protected characteristic per federal, state, or local law.
#J-18808-Ljbffr
In Summary: Swarmbotics AI is seeking a Machine Learning Engineer to design, develop, and deploy a multi-modal action model that enables each UGV to select and execute coordinated swarm macro-actions in real time . This role sits at the intersection of machine learning and multi-agent decision making . The core focus is on the decision-making and action-selection layers .
En Español: La empresa Swarmbotics AI es una compañía de robótica en enjambre de bajo costo para la industria y la defensa. Vemos un mundo ubicuo de robots de bajo coste que transforman casi todos los aspectos de la sociedad, pero vemos una necesidad urgente en la industria defensiva. Nos centramos en construir enjambres de robots que incorporan un BOM de bajo precio, una pila autónoma optimizada para componentes fuera del estante y un planificador global que permite capacidades de enjambre para grupos de robots para realizar tareas sofisticadas. Nuestro primer producto es una aplicación de defensa construyendo vehículos no tripulados terrestres (UGVs), colectivamente denominados - Attritable, Network Founded, Tactical Swarm (ANTS). Cada UGV en ANTS es un robot de acción ajetreable independiente diseñado para manejar a demanda y movilidad humana. Este papel se encuentra en la intersección del aprendizaje automático y la toma de decisiones multi-agentes: construirás modelos aprendidos que razonan sobre entradas multimodales para realizar macroacciones tácticas. Trabajará en estrecha colaboración con el liderazgo de la empresa y los equipos interfuncionales para alinear las capacidades con la hoja de ruta del producto Swarmbotics AI. What You'll Do Arquitecto, entrenamiento e iteración sobre modelos de acción multimodales que seleccionan macroacciones tácticas a nivel de enjambre desde fuertes entradas contextuales Diseñar arquitecturas de modelo que fusionan insumos heterogéneos percepción local, estado de enjambre, objetivos de misión en una representación unificada de decisiones Desarrollar y aplicar enfoques de aprendizaje de refuerzo online y offline, incluyendo modelado de secuencias basado en transformadores, aprender políticas de coordinación conjunta Optimizar modelos para ejecutar dispositivos en tiempo real a través de cuantización, destilación y eficiencia Conceptos de diseño de maquinaria de construcción y desarrollo de procesos más eficaces A partir de la recopilación de datos por medio de la tensión conjunta Tensión de trabajo mediante capacitación, evaluación y análisis de tareas de ejemplares Los propietarios de un proyecto de trabajo en línea desarrollan y aplican estrategias de desarrollo automático de sistemas de gestión de recursos humanos (PLAV) Optometría de proyectos o métodos de formación autónomos como la generación de conocimientos básicos y implementación de software, planificación de programas de programación de trabajo, planeación multitécnicas y configuración de computadoras y redes de Python, Experiencias avanzada o aplicaciones técnicas de aprendizajes, etc. Swarmbotics es un empleador de igualdad de oportunidades. Todos los solicitantes calificados serán tratados con respeto y recibirán la misma consideración para el empleo sin importar raza, color, casta, credo, religión, sexo, identidad de género, orientación sexual, origen nacional, ascendencia, discapacidad, servicio uniforme, estatus de veterano, edad o cualquier otra característica protegida por ley federal, estatal u local. #J-18808-Ljbffr