Research Engineer, Robotics Learning - Sunnyvale, United States - 1X Technologies Cable Company

Description

Targeted start date:
Immediately. Relocation provided.

We have four specializations under the role of

Research Engineer:
Foundation Models, Infrastructure, Simulation, and Reinforcement Learning. Please specify your preferred specialization when applying for this position.

1X's mission is to create an abundant supply of physical labor through safe, intelligent androids that work alongside humans.

Since 2014, we've designed our products with large-scale manufacturing in mind, so we can produce androids at a high enough volume to meet the world's labor demand.

Our wheeled Android EVE, is engineered to work with you, from guarding to logistics, and our bipedal model, NEO, is designed to become a household android with broad deployment in various applications.

Our products will understand both natural language and physical space, so they can complete useful tasks in any environment.

We've established our dual HQ in San Francisco and Norway. Our positions require in-person presence to ensure effective execution and seamless collaboration in our hardware-focused environment. We value passion for our work and encourage candidates who share our dedication to join our team.

Foundation Models

Research Engineers in foundation models are generalists who build and maintain end-to-end responsibility for our AI models.

A typical day involves implementing new model architectures, developing full-stack infrastructure for ramping up our data engine, or implementing new ways to evaluate general-purpose robot policies.

You will train and scale deep neural networks for manipulation, navigation, and locomotion on real robot hardware and improve both the breadth of skills and level of success of our androids.

You will work closely with other AI team members and also develop independently.

Why this job is exciting (Foundation Models)

• Your role on the team will be to scale up learning algorithms on state-of-the-art hardware
• You will build upon one of the largest android datasets, which is continuing to grow exponentially
• We aim to (1) double the number of tasks the robot can perform and (2) half the error rate on existing tasks, quarter over quarter
• The team works closely together to prioritize direction and scale up a single approach to a production-ready system

Responsibilities (Foundation Models)

• Day to day: implement and scale up models for end-to-end navigation and manipulation and locomotion
• Build the data engine (frontend UI and backend) to log, clean, and label data for foundation model training
• Solve the general-purpose evaluation problem and bring robot skills to 99%+ success
• Train and evaluate policies in simulation and solve sim-to-real
• Work with our robot operations team to scale up our datasets and model capabilities

Must-Haves (Foundation Models)

• Bachelor's degree in Computer Science or equivalent
• Published research in top ML conferences (NeurIPS, CVPR, ICML, ICLR, CoRL, RSS, etc.)
• Proficiency working with and testing large codebases in Python
• Ability to rapidly prototype ideas in code in an independent manner
• Familiarity with linear algebra and supervised machine learning
• Experience with Deep Learning Frameworks (Pytorch, TF, JAX, etc.)

Nice-to-haves (Foundation Models)

• ROS/ROS2 experience
• Have built large programming projects on your own for fun (e.g. open source projects)
• Training large-scale ML models such as visual foundation models, large language models, pixel-level generative models
• You spend time reading r/LocalLLaMA/ and tinkering with open source LLM technology
• Have implemented highly performant software

Infrastructure

Research Engineers in infrastructure are generalists who build and maintain infrastructure and platforms for robot learning.

A typical day involves scaling up data and compute engines, or improving the reliability and uptime of model training and inference.

You will be responsible for scaling model training and deployment, data engines and databases, and frontends and backends. You will work closely with other AI team members and also develop independently. Success in this role means unblocking data pipelines and compute scaling for an exponentially growing fleet of robots.

Why this job is exciting (Infrastructure)

• Your role on the team will be to scale up the learning stack on state-of-the-art hardware and drive success of robot operations
• You will build upon one of the largest android datasets, which is continuing to grow exponentially
• We aim to (1) double the number of tasks the robot can perform and (2) half the error rate on existing tasks, quarter over quarter
• The team works closely together to prioritize direction and scale up a single approach to a production-ready system

Responsibilities (Infrastructure)

• Day to day: ship AI platform for non-technical robot operators, scale up AI model training workloads, and own model inference reliability
• Build the data engine (frontend UI and backend) to log, clean, and label data for foundation model training
• Work with our robot operations team to scale up our datasets and model capabilities

Must-Haves (Infrastructure)

• Bachelor's degree in Computer Science or equivalent
• 5+ years of professional, full-time experience building ML infrastructure
• Proficiency working with and testing large codebases in Python, and C++ or Rust
• Experience with databases (Postgres, NoSQL)
• Ability to rapidly prototype ideas in code in an independent manner
• Experience with Deep Learning Frameworks (Pytorch, TF, JAX, etc.)
• Ability to build scrappy, cost-effective solutions to prove out ideas quickly and then harden them for production use cases

Nice-to-haves (Infrastructure)

• Experience managing ML compute clusters, cloud infrastructure, and workload management, and setting up hybrid cloud infra
• Have implemented highly performant software
• Have built large programming projects on your own for fun (e.g. open source projects)
• Training large-scale ML models such as visual foundation models, large language models, pixel-level generative models

Simulation

Research Engineers in simulation are generalists who build the simulation stack and close the sim-to-real gap.

A typical day involves building contact-rich physics simulation, scaling up reinforcement learning algorithms, or solving the sim-to-real gap in the visual or physics domains.

You will own simulation and focus on making simulation as widely used as possible within the company. By making simulation a trusted proxy of the real world, you will drive AI model evaluation and iteration speed. You will work closely with other AI team members and also develop independently.

Why this job is exciting (Simulation)

• Your role on the team will be to build ML robotics simulation and close the sim-to-real gap
• You will work across the AI stack and solve gaps in model training and inference
• We aim to (1) double the number of tasks the robot can perform and (2) half the error rate on existing tasks, quarter over quarter
• The team works closely together to prioritize direction and scale up a single approach to a production-ready system

Responsibilities (Simulation)

• Day to day: ship AI stack for non-technical robot operators, scale up AI model training workloads, and own model inference reliability
• Bring real-world tasks and projects into simulation via real-to-sim
• Build the data engine (frontend UI and backend) to log, clean, and label data for foundation model training
• Work with our robot operations team to scale up our datasets and model capabilities

Must-Haves (Simulation)

• Bachelor's degree in Computer Science or equivalent
• 3+ years of professional, full-time experience building simulation
• Experience with physical simulators (Pybullet, IssacSim, Mujoco, etc.) and training policies in simulation
• Experience with physics simulation (contact dynamics, system identification, physics engines)
• Experience learning-based approaches to close the sim-to-real gap (domain adaptation, randomization)
• Experience generating simulated assets and environments (Blender, Maya)
• Ability to rapidly prototype ideas in code in an independent manner
• Published research in top Simulation or ML conferences (SIGGRAPH, NeurIPS, CoRL, RSS, etc.)

Nice-to-haves (Simulation)

• Experience with high-fidelity, real-time rendering
• Experience with GPU simulation
• Have built large programming projects on your own for fun (e.g. open source projects)
• Training large-scale ML models such as visual foundation models, large language models, pixel-level generative models
• Have implemented highly performant software

Reinforcement Learning

Research Engineers in reinforcement learning are generalists who develop reinforcement learning algorithms for humanoid locomotion and manipulation policies. A typical day involves prototyping reinforcement learning algorithms or building distributed infrastructure for training and simulation. You will own the RL policies and train in both sim and real to bring capabilities to NEO. You will integrate and distill RL policies and data into foundation model training. You will work closely with other AI team members and also develop independently.

Why this job is exciting (Reinforcement Learning)

• Your role on the team will be to build RL algorithms in simulation and real and integrate RL into the overall learning picture
• You will work across the AI stack and solve gaps in model training and inference
• We aim to (1) double the number of tasks the robot can perform and (2) half the error rate on existing tasks, quarter over quarter
• The team works closely together to prioritize direction and scale up a single approach to a production-ready system

Responsibilities (Reinforcement Learning)

• Day to day: build RL algorithms to bring generalized capabilities to NEO
• Work with our robot operations team to scale up our datasets and model capabilities

Must-Haves (Reinforcement Learning)

• Bachelor's degree in Computer Science or equivalent
• 3+ years of professional, full-time experience on robot learning
• Experience with physical simulators (Pybullet, IssacSim, Mujoco, etc.) and training RL policies in simulation
• Ability to rapidly prototype ideas in code in an independent manner
• Published research in top ML conferences (NeurIPS, CoRL, RSS, ICML, etc.)

Nice-to-haves (Reinforcement Learning)

• Experience with GPU simulation
• Have built large programming projects on your own for fun (e.g. open source projects)
• Training large-scale ML models such as visual foundation models, large language models, pixel-level generative models
• Have implemented highly performant software

#J-18808-Ljbffr