“AI robotics” became one of the most-discussed venture sectors in 2024–2025 with billions invested in humanoid and embodied-AI companies. By 2026, several have product traction or investor commitments significant enough to make hiring real. Interviews are different from typical software roles — they blend ML, hardware, simulation, control theory, and operational reality. This guide is for the engineer evaluating the move.
The companies hiring
- Figure: humanoids for industrial and home use; OpenAI partnership for cognition
- 1X (formerly Halodi): Norwegian humanoid; deep ML team
- Boston Dynamics: the established player (Spot, Atlas, Stretch); Hyundai-owned
- Tesla Optimus: integrated with Tesla’s production manufacturing
- Agility Robotics: Digit, focused on warehouse work
- Sanctuary AI: Canadian, AGI-focused humanoids
- Apptronik: Apollo, U.S.-based humanoid
- Specialty roboticists: Covariant (manipulation), Skydio (drones), Anduril (defense)
What the interviews probe
- Strong CS fundamentals (DSA, systems)
- ML literacy — especially RL, imitation learning, computer vision
- Robotics-specific knowledge — kinematics, control, simulation
- Practical sim-to-real intuition
- Hardware empathy — what real robots actually do vs simulation
Typical interview shape
- Recruiter screen with technical depth probing
- Coding round (Python with NumPy / PyTorch)
- Domain round: depending on role — control, perception, planning, learning
- System design — robotics-flavored (autonomous task pipeline)
- Craft / project deep-dive
- Behavioral with mission-fit lens
Domain rounds — pick your specialty
Perception
- 3D scene understanding
- SLAM (Simultaneous Localization and Mapping)
- Object detection and grasp planning
- Coordinate transforms and camera intrinsics
Control
- Inverse kinematics and dynamics
- Model-predictive control (MPC)
- Task-space vs joint-space control
- Haptic / force-feedback handling
Learning
- Reinforcement learning (PPO, SAC)
- Imitation learning (behavioral cloning, DAgger)
- Foundation models for robotics (RT-2, OpenVLA, etc.)
- Sim-to-real transfer techniques
Planning
- Motion planning (RRT, A*, trajectory optimization)
- Task planning (PDDL, hierarchical decomposition)
- Multi-step task execution
- LLM-driven planning (the 2024–2026 wave)
The simulation question
“Tell me about your sim-to-real workflow.” Strong answers cover:
- Simulator (Isaac Gym, MuJoCo, PyBullet)
- Domain randomization to bridge sim-to-real gap
- Real-world data collection for policy fine-tuning
- Failure modes when policies transfer poorly
System design
Common prompts:
- “Design a pipeline for a humanoid that picks an item off a shelf”
- “Design a teleoperation data-collection system for behavior cloning”
- “Design the safety architecture for a humanoid in a home”
- “Design the over-the-air update system for a fleet of robots”
The safety probe
Robotics interviews probe safety thinking heavily:
- What watchdogs and emergency stops are required?
- How do you bound failures of a learned policy?
- What is your testing strategy for safety-critical changes?
- How do you handle the human-robot interaction edge cases?
Skills to brush up
- Linear algebra and rigid-body kinematics
- PyTorch / JAX for ML
- ROS or ROS 2 (still common)
- Isaac Sim / Isaac Lab if relevant to the company
- One real implementation in your background — even a small one
Compensation
- Senior IC: $200K–$350K total cash; equity heavy at early stage
- Staff: $300K–$500K total at established companies; $400K+ at hot startups
- Specialty roles (e.g., control with dexterous-manipulation experience): premium
- Boston Dynamics, Tesla compensate at parent-company tier (Hyundai-tied / Tesla)
- Figure, 1X, etc.: VC-funded startup pay (cash + significant equity)
The honest case for the move
- Most interesting AI work happens at robotics companies in 2026
- Hardware-software integration is genuinely fun
- Equity upside if a winner emerges
- Strong technical colleagues
The honest case against
- Many companies will not survive (capital-intensive market)
- Hardware iteration is slower than software (years vs months)
- Production reality is harsh (safety, regulatory)
- The “humanoids in homes” thesis remains unproven
How to break in
- Build a robotics side project (sim or low-cost real robot)
- Contribute to ROS, Isaac Lab, MuJoCo
- Read foundational papers (RT-2, OpenVLA, MobileALOHA)
- Take a robotics MOOC (Berkeley, Stanford robotics curricula)
- For ML specialists: focus on imitation learning, RL, foundation models
What separates senior candidates from junior
Junior candidates know individual concepts. Senior candidates discuss the integrated stack — perception → planning → control → execution → feedback — and the operational realities (safety, certification, cost). Staff-level discussion includes the company-strategy lens (which problems are tractable, which are 5+ years out).
Frequently Asked Questions
Do I need a robotics degree?
Not strictly. Strong ML or systems background plus public robotics work usually suffices. Specific roles (manipulation control) prefer formal training.
Is “humanoid robots in homes” really happening?
Demos are improving rapidly. Reliable home deployment in 2026 is still rare; warehouse / industrial is the immediate market. The home thesis is multi-year.
Should I join early-stage or established?
Early-stage (Figure, 1X) for equity upside and varied work; established (Boston Dynamics) for stable comp and proven products. Pick by risk tolerance.