Backend System Team
Design, build, and own the core backend systems that power our industrial robotics platforms.
Team Overview
The Backend System team designs, builds, and owns the core backend systems that power our industrial robotics platforms.
We take end-to-end ownership of backend components supporting robot controllers, working close to hardware, sensors, and operating systems to deliver reliable, high-performance systems in real-world production environments. The team designs and operates low-latency services, data-intensive pipelines, and robust communication paths that directly influence system behavior on the factory floor.
Our code doesn’t just live on servers. It connects different functions and layers of intelligence, enabling real industrial robots to perceive, plan, and operate in the real world.
Backend code runs on the controller inside every robot cell: it drives the robot arms, feeds 3D camera data into perception, talks to PLCs and conveyors, and links each controller to the fleet for provisioning and upgrades.
If it fails, a cell stops and the line halts, a stale state causes a mispick or dropped item, support goes blind without telemetry, or a failed upgrade takes a controller offline at a customer site with no engineer nearby.
Products & Applications Powered by This Team
Mujin Robot Controller
Powers low-latency service architectures, real-time inter-process communication, and sensor ingestion pipelines driving perception and motion execution inside every production robot cell.
Mujin Portal & Remote Maintenance
Orchestrates edge-to-cloud diagnostic telemetry, unified incident timelines, and safe, automated software provisioning across global heterogeneous controller fleets.
Tech Stack
Languages & Runtimes
- Modern C++
- Python 3.11+
- Go
- Linux
Key Libraries, Frameworks & Math
- Messaging and serialization libraries, async networking, API frameworks, relational and in-memory data stores.
- Core focus: concurrency, async I/O, inter-process communication, and data pipelines.
Real-Time IPC, Fieldbuses & Protocols
- EtherCAT
- HTTP / GraphQL / WebSocket
- ZeroMQ IPC
Hardware, Testbeds & Tools
- Tokyo HQ testing floor with production robot cells
- Multi-vendor 3D cameras and industrial sensors
- In-house simulation / digital twin environment
- Docker
- Profiling and debugging tooling
- In-house controller diagnostics tooling
Technical Challenges
Safe upgrades across a heterogeneous controller fleet
Challenge: Controllers run 24/7 at customer sites across several hardware and software generations. Changing software on a live production system carries downtime risk, with no engineer on site.
Approach: Made every deployed release reproducible and traceable, automated the upgrade path with verification before and after, and built visibility into what is actually running on each controller.
Impact: Upgrades reach controllers of every generation without site visits. Surprises are found before an upgrade rather than after.
Multi-vendor 3D camera integration, testable without cameras
Challenge: Each camera vendor behaves differently in interface, timing, and failure modes. Regressions could only be caught on physical hardware, so bugs reached customer sites late. Capture also has a tight per-cycle time budget.
Approach: Unified vendor differences behind one abstraction and made the full sensor path runnable in simulation, so the same code is tested automatically on every change and simulation stays faithful to real hardware.
Impact: Regressions caught before hardware is involved. New vendors onboard behind the same interface. Timing regressions surface in simulation.
Diagnosing production issues on remote controllers
Challenge: Controllers sit on customer networks with limited access. Diagnostic data is scattered across many processes and differs between controller generations, so support engineers lacked a coherent picture.
Approach: Built a single diagnostics interface that assembles a coherent timeline of a controller’s behavior across sources and generations, safe to use on production systems.
Impact: Incident timelines reconstructed in minutes instead of hours. Root causes found without site visits.
Typed, low-latency controller data layer
Challenge: Controller state crossed language and component boundaries without enforced contracts, so breakage surfaced at runtime. Consumers saw stale state, and API performance degraded under load.
Approach: Introduced enforced data contracts shared across all languages, moved state delivery from polling to push, and reworked the data access layer for load. Kept remote sites reachable regardless of network setup.
Impact: Contract breaks caught at build time. Consumers see state as it changes. API latency stays flat under load and remote sites stay reachable.
Key Responsibilities
- Architect and own backend components of the robot controller end-to-end, from design to production.
- Design low-latency services and IPC paths that shape controller behavior on the factory floor.
- Integrate 3D cameras and industrial devices; debug across application logic, OS, drivers, and device interaction.
- Optimize high-performance data pipelines, database access patterns, and caching.
- Build and maintain the controller’s API surface and services with emphasis on stability and performance.
- Diagnose production issues across system layers, improve observability tooling, and support provisioning and upgrade workflows.
Weekly Engineering Rhythm
- 60% Design and software code
- 25% Simulation / digital twin / bench test
- 15% Hands-on testing with real robots / hardware in Tokyo HQ
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:
- Enjoy taking ownership of challenging, ambiguous problems and driving them forward proactively.
- Curious and motivated to dig deeply into unfamiliar problems and figure things out rather than waiting for a ready-made answer.
- Enjoy working across the broader robot controller stack rather than being limited to one narrowly defined component.
- Value being able to investigate issues directly in the actual robot and hardware environment when needed.
- Thrive on collaborating across teams and seeing their work have an impact beyond their immediate area.
Who Might Struggle on This Team
- Engineers who need narrow, fully-specified tickets before they can start.
- Engineers who prefer pure cloud / web work and avoid hardware or on-site debugging.
- Engineers who treat the OS, drivers, and network as a black box when things break.
- Engineers who prioritize feature velocity over 24/7 production reliability.
Join the Team
Qualifications & Growth
- Advanced C++ and Python.
- Strong Linux internals, networking, and async I/O.
- Docker.
- Full-stack development background.
- English technical communication.
- BSc in Computer Science or equivalent experience.
- EtherCAT and real-time Linux (PREEMPT_RT, kernel-adjacent work).
- Multi-vendor 3D camera integration.
- Robot controller architecture: how planning, vision, and I/O modules compose over IPC.
- Fleet operations: provisioning, upgrades, and observability for large on-premises fleets.