Apptronik · Austin, TX, USA, United States
The Role
Lead development and deployment of state-of-the-art RL algorithms for locomotion and manipulation on humanoid robots. Prototype in simulation, transfer and fine-tune on hardware, optimize distributed training pipelines, mentor junior engineers, and collaborate with hardware teams to analyze results and improve sim-to-real performance.
Summary Generated by Built In
Apptronik is a human-centered robotics company developing AI-powered robots to support humanity in every facet of life. Our flagship humanoid robot, Apollo, is built to collaborate thoughtfully with people, starting with critical industries such as manufacturing and logistics, with future applications in healthcare, the home, and beyond.
We operate at the cutting edge of Applied AI, applying our expertise across the full robotics stack to solve some of society's most important problems. You will join a team dedicated to bringing Apollo to market at scale, tackling the complex challenges like safety, commercialization, and mass production to change the world for the better.
JOB SUMMARY
The Senior Reinforcement Learning Engineer is a key, hands-on role focused on achieving state-of-the-art performance on our humanoid robots. This engineer will leverage their deep expertise in RL to solve critical locomotion and manipulation challenges and deliver breakthrough results on physical hardware. The primary focus of this role is to rapidly implement, iterate, and deploy advanced learning algorithms to push the boundaries of what our robots can do. As a senior member of the team, this individual will also be responsible for mentoring junior engineers, elevating the team's overall technical capabilities through their guidance and expertise.
ESSENTIAL DUTIES AND RESPONSIBILITIES or KEY ACCOUNTABILITIES
Implement and deploy state-of-the-art RL algorithms to achieve ambitious, world-class performance on dynamic locomotion and manipulation tasks with physical hardware.
Drive the entire development cycle, from prototyping in simulation to robustly transferring and fine-tuning policies on the robot.
Optimize and scale the RL training pipeline for faster iteration, contributing to core infrastructure for high-throughput simulation and distributed training.
Mentor junior engineers by providing technical guidance, conducting insightful code reviews, and sharing best practices in reinforcement learning and software development.
Collaborate closely with the robotics and hardware teams to diagnose system-level issues and improve sim-to-real performance.
SKILLS
5+ years hands-on expertise with RL frameworks (PyTorch, JAX) and high-fidelity physics simulators (MuJoCo, IsaacGym)
Mastery of Python for rapid prototyping and strong proficiency in C++ for performant, deployable code
Experience building or utilizing large-scale, distributed training pipelines and optimizing them for throughput
Strong theoretical understanding of modern reinforcement learning, including imitation learning, model-based RL, and sim-to-real transfer
Strong intuition for robot dynamics and controls theory to guide and constrain learning-based approaches
Proven track record of successfully deploying learning-based policies on physical robotic systems (legged robots or manipulators)
Demonstrated experience mentoring or providing technical guidance to other engineers
Ability to develop and refine motion retargeting pipelines using human demonstration data (mocap, teleoperation)
PhD or MS in Computer Science, Robotics, or related field (2+ years industry experience strongly preferred)
Strong publication record in relevant conferences or journals (CoRL, RSS, ICRA) — Preferred
LANGUAGE REQUIREMENT: Detect the primary language used in the source document and write all generated descriptive text fields in that SAME language. For JD this includes: description_summary, intro_text, responsibilities, generated_html_description. Return salary fields separately as requested: salary_type, salary_period, salary_currency, salary_min, salary_max. Do NOT translate these fields to English unless the source is English. Detected source language code: en (English). Use this detected language for all generated narrative fields. If uncertain, default to English (en).