Research Engineers & Research Scientists — Robotics & Physical AI

San Francisco · on-site

PerfectBit is building technology to help robots learn, reason, and operate in the physical world.

Robotics requires advances across algorithms, simulation, hardware, and real-world experimentation. Robots need to perceive their environments, understand tasks, plan actions, and improve through interaction. Building capable robotic systems requires breakthroughs in learning, control, simulation, and the infrastructure that connects them.

PerfectBit develops AI-driven methods for advancing robotics. Our work may include:

• Developing learning systems for robots, including policies, world models, and vision-language-action models
• Creating simulation environments and evaluation frameworks for robotic tasks
• Applying and developing advanced computer vision and perception algorithms for autonomous robots
• Building systems that connect simulation and real-world robot deployment
• Working with robot hardware, sensors, and platforms to develop and test new capabilities
• Using synthetic and real-world experiences to improve robotic performance
• Creating tools and infrastructure that accelerate robotics research and development

We aim to infuse cutting-edge AI research with practical robotics engineering to deliver robot policies that are measurably better.

About the team

• Peter Vajda was previously Director of Media Generation at Meta Superintelligence Labs, where he led the foundation-model teams behind Movie Gen and Emu. He was previously a Visiting Assistant Professor at Stanford and holds a PhD in computer science.

• Seiji Yamamoto was previously a Senior Staff Research Scientist at Meta Superintelligence Labs, where he led teams in the Core Llama organization spanning LLM pre-training and post-training, inference, speech, and vision. He holds a PhD in physics.

What You Will Work On

You will help build systems that advance the capabilities of robots and embodied models.

We are hiring for several roles. Depending on your background, your work could include:

• Building reinforcement-learning and imitation-learning systems for humanoids, manipulation, locomotion, or mobile robotics
• Developing simulation environments in Isaac Lab, MuJoCo, MJX, Genesis, or related platforms
• Training and evaluating vision-language-action models, world models, and robot policies
• Developing real-to-sim, sim-to-real, and real-to-sim-to-real pipelines
• Working directly with robotic hardware, sensors, actuators, and embedded systems
• Integrating and experimenting with robot platforms to validate new algorithms in the physical world
• Creating evaluation frameworks and benchmarks for measuring robot capabilities
• Working directly with robotics companies and research labs to solve challenging technical problems

This is an early-stage role. You will influence both our technical direction and the products we build.

Who we’re seeking

We are looking for people who can operate with substantial autonomy and turn ambiguous research problems into working systems.

You may be a strong fit if:

• You have built substantial robotics, machine-learning, simulation, hardware, or infrastructure systems
• You are comfortable moving between research, engineering, experimentation, and product development
• You can identify important problems, scope practical solutions, and ship
• You care more about measurable technical progress than organizational process
• You work well in a small, highly collaborative team
• You use AI coding tools such as Claude or Codex to increase your speed and scope
• You enjoy working directly with robots, hardware platforms, simulators, and real-world experiments
• You want to build new methods rather than apply a fixed playbook

Location

• On-site in San Francisco (this role is not remote)

Compensation

•  $100K–$300K base + meaningful equity

1. Informal meeting
2. Technical interview
3. Behavioral Interview
4. Candidate presentation on technical project(s) + Q&A