Job description
Help us build CloudWalk’s ambitious R&D program around humanoid robotics. You’ll own the full cycle of developing locomotion and high-level behaviors: design tasks and environments in NVIDIA IsaacLab/Sim/Gym, train policies with RL/IL, validate on simulation and deploy on our physical humanoid robot (Sim2Real).
The Robotics Division
- We work on embodied AI and autonomous systems across various robotics platforms and applications
- Full development cycle from simulation environment design in NVIDIA IsaacLab/Sim/Gym to deployment on physical robots
- São Paulo lab with advanced robotics hardware including humanoids and vending machines
What you’ll do:
- Invent and build humanoid tasks and environments – from simple flat-ground locomotion to challenging obstacle courses
- Train and refine policies using RSL-RL, RL-Games, Stable-Baselines3, or similar GPU-parallel RL libraries
- Explore cutting-edge approaches – reinforcement learning, imitation learning, diffusion policies, and transformer-based approaches for continuous control
- Close the Sim2Real gap through domain/dynamics randomization, calibration, and staged deployment on the humanoid platform in our lab
- Integrate with robotics ecosystem – ROS 2 (ros2_control, MoveIt 2) and NVIDIA’s robotics stack (Isaac ROS, cuRobo/cuMotion, NVBlox)
- Orchestrate complex behaviors with BehaviorTree.CPP/Groot, and experiment with Isaac GR00T for high-level humanoid skill composition
- Measure and iterate with reproducible pipelines, dashboards, and design documentation to accelerate R&D cycles
What you’ll need:
- Solid ML and RL foundations (PPO, SAC, TD3) with hands-on PyTorch experience
- NVIDIA Isaac ecosystem proficiency – Isaac Sim/Isaac Lab and legacy environments (Isaac Gym, OmniIsaacGymEnvs, Orbit)
- Strong ROS 2 skills – ros2_control, lifecycle nodes, planning and perception integration
- Sim2Real expertise – domain randomization, dynamics tuning, and safe hardware deployment
- Engineering best practices – version control, containerization, reproducibility, and documentation
- Must be able to work on-site in São Paulo for regular robotics lab sessions with physical robots
Bonus points if you have:
- Legged locomotion experience with humanoid control tasks (Legged Gym, ETH Zurich RSL environments)
- MuJoCo-based RL workflows (Gymnasium, dm_control, robosuite) for algorithm prototyping and benchmarking
- Advanced policy architectures – Diffusion/Transformer policies for robotics applications
- NVIDIA robotics stack – GPU-accelerated planning (cuRobo, cuMotion) and Isaac ROS perception (NVBlox, VSLAM)
- Real-time systems and mechatronics knowledge (PREEMPT_RT, C++ optimization)
- Strong portfolio of robotics and ML projects (GitHub repos, demonstrations, research papers)
Work Model
- Hybrid role – collaborate remotely but spend regular days in our robotics lab
- Hands-on experimentation with physical humanoid robots is a core part of this job
- Test, validate, and deploy your simulation work on real hardware in our São Paulo lab
Recruiting Process
- Challenge: You’ll receive a technical challenge to demonstrate your skills in robotics, simulation, and machine learning
- Technical interview
- Cultural interview
Note: If you are not willing to take an online quiz and demonstrate your technical capabilities with both simulation and physical robotics systems, do not apply.
Diversity and inclusion:
We believe in social inclusion, respect, and appreciation of all people. We promote a welcoming work environment, where each CloudWalker can be authentic, regardless of gender, ethnicity, race, religion, sexuality, mobility, disability, or education.
