Robotics System Team
Core real-time motion planning, dynamic inverse kinematics, and collision-free trajectory optimization for autonomous industrial robots.
Team Overview
The Robotics System team designs, implements, and optimizes the algorithmic core of MujinOS. Our primary mission is solving the fundamental challenge of autonomous manipulation: transforming high-level task goals (e.g. “depalletize this mixed SKU pallet”) into smooth, collision-free, dynamic trajectories executed on multi-vendor 6-DOF and 7-DOF industrial robot arms in real time.
Our software platform replaces traditional manual “teach pendants” with autonomous motion generation. Operating at the intersection of computational geometry, Lie group theory, and real-time control, the Robotics System engine computes inverse kinematics, avoids singular configurations, and dynamically coordinates multi-axis industrial robots within strict sub-millisecond control loop budgets.
Products & Applications Powered by This Team
Mujin Robot Controller
Powers real-time trajectory generation, collision avoidance, and pick-and-place execution across Depalletizing, Palletizing, Piece Picking, and Bin Picking.
Real-Time Digital Twin
Provides continuous pre-flight kinematic reachability checks, singular point detection, and multi-robot workspace coordination in virtual simulation.
Tech Stack
Core Languages & Execution
- Languages: C++ (low-latency kinematics solver), Python 3.11+ (task coordination & planning harness)
- Environment: Real-Time Linux (
PREEMPT_RT), Ubuntu LTS - Build Toolchains: CMake, Docker
Kinematics & Algorithmic Engines
- Motion Planning & IK: MujinMI motion planning engine, closed-form analytical inverse kinematics solvers (IKFast)
- Math Foundations: Eigen3, custom Lie algebra implementations ($SO(3)$, $SE(3)$), spatial transformation matrices
- Geometric Reasoning: BVH (Bounding Volume Hierarchy) tree acceleration, continuous collision checking in configuration space ($C$-space)
Real-Time Concurrency & IPC
- Inter-Process Communication: Lock-free ring buffers, POSIX shared memory (
shm_open), ZeroMQ - Serialization: Protocol Buffers, FlatBuffers for zero-copy deserialization
- Driver Interfaces: Low-latency fieldbus adapters communicating with Fanuc, Yaskawa, ABB, Kawasaki, and KUKA robot controllers
Simulation & Verification
- Physical Testbed: Tokyo HQ Robotics Lab with 50+ industrial manipulators (Fanuc, Yaskawa, ABB, Kawasaki, KUKA)
- Digital Twin: WebGL real-time 3D simulation with dynamic obstacle modeling, verifying kinematic reachability and collision clearance
- Continuous Testing: Automated Hardware-in-the-Loop (HIL) test cells with physical manipulators
- Analysis Tooling: AddressSanitizer (ASan), ThreadSanitizer (TSan), Perf, Tracy Profiler
Technical Challenges
Real-Time Dynamic Obstacle Avoidance
Challenge: When obstacles enter the cell or dynamic environment states change mid-motion, trajectory recalculation previously risked triggering protective safety stops or operational pauses.
Approach: Implemented real-time dynamic trajectory recalculation in high-dimensional configuration space ($C$-space) running on an isolated real-time CPU core, updating trajectory setpoints on the fly.
Impact: Enables smooth dynamic detour around obstacles at tool tip velocities up to 2.5 m/s without stopping the robot or halting factory production throughput.
Vibration Suppression & Cycle-Time Optimization
Challenge: Moving delicate, heavy, or variable payloads under aggressive industrial cycle times causes mechanical resonance, gearbox stress, and payload instability.
Approach: Formulated an optimal multi-phase trajectory optimization algorithm with dynamic acceleration-profile shaping and real-time corner rounding.
Impact: Reduced end-effector mechanical vibrations by 42% while improving overall pick-and-place cycle times by 8%.
Key Responsibilities
- Dynamic Trajectory Generation: Design trajectory optimization algorithms that maximize throughput (speed and acceleration limits) while respecting motor torque curves and mechanical constraints.
- Singularity & Workspace Management: Develop robust mathematical strategies to identify, avoid, and navigate kinematic singularities in 6-DOF/7-DOF manipulators without halting execution.
- Real-Time Collision Avoidance: Maintain sub-millisecond continuous collision queries against dynamic point clouds and geometric obstacles streaming from the Computer Vision pipeline.
- Hardware Integration & Driver Bridges: Develop deterministic communication layers interfacing with industrial robot servo controllers across proprietary vendor protocols (Fanuc, Yaskawa, ABB, Kawasaki, KUKA).
- Digital Twin & HIL Test Verification: Build deterministic integration simulations to validate pick-and-place cycles across thousands of virtual edge cases before deploying to live factory sites.
- Production Reliability & Telemetry: Directly monitor telemetry from thousands of robot operating hours across global fulfillment and manufacturing hubs, diagnosing edge cases and deploying zero-downtime hotfixes.
Who Will Enjoy and Excel on This Team
If you want your code to control the physical world rather than just move pixels on a screen, this team is an incredible place to be. You will excel and have the most fun here if you:
- Love Seeing Code Move Real Machinery: You get a genuine rush when your code does not just write database records, but commands a 1,000 kg industrial robot moving at 2.5 m/s with millimeter accuracy.
- Enjoy Solving Problems from First Principles: When a robot motion fails or hits a torque limit, you do not guess. You genuinely enjoy opening kinematics models, coordinate transforms, and physics calculations to understand the root cause.
- Take Pride in High-Performance Code: You care deeply about cache efficiency, zero-allocation real-time loops, thread safety, and writing deterministic C++ code that runs reliably 24/7.
- Like Hands-on Testing (The Gemba): You prefer walking down to the testing floor and debugging on real robot hardware over spending all day in meetings or on theoretical designs.
- Want High Autonomy in a Global Team: You thrive when given the freedom to own architectural choices, test new ideas quickly, and collaborate with passionate engineers from 26+ countries.
Open Roles & Requirements
Required Skills vs. What You Can Learn
We look for strong first-principles thinkers rather than domain perfectionists. You do not need to check every box on day one.
- Solid C++ Proficiency: Comfortable with C++, pointer safety, memory management, and performance debugging.
- Applied Mathematics: Firm grasp of linear algebra, vector spaces, coordinate transforms, and basic 3D geometry.
- Data Structures & Algorithms: Experience with graph traversal, computational geometry fundamentals ($A^*$, KD-trees, collision bounds).
- Systems Curiosity: Enthusiasm for debugging low-level concurrency, race conditions, and real-time execution constraints.
- Lie Group & Screw Theory: High-order $SO(3)$ and $SE(3)$ kinematics representation and dynamic singularity avoidance.
- Real-Time OS Internals: Linux
PREEMPT_RTscheduling policies, memory locking (mlockall), and CPU affinity tuning. - Industrial Robot Hardware: Proprietary driver interfaces for Fanuc, Yaskawa, ABB, Kawasaki, and KUKA robot controllers.
- Multi-Agent Coordination: Complex dual-arm synchronization and high-speed dynamic trajectory interpolation.
Interview Process
This team evaluates candidates through our standardized Algorithmic & Systems Track:
- Recruiter Screen (30 min): Background, motivations, and culture alignment.
- Technical Coding & Math Assessment (60–90 min): Algorithmic problem solving, data structures, and spatial geometry intuition.
- System Architecture Deep Dive (60 min): Discussion on motion planning tradeoffs, concurrency, and real-time performance.
- Gemba Tour & Leadership Conversation (60 min): Live tour of Tokyo HQ robot testing cells and mutual expectation alignment.