Robotics control engineer · Shenzhen

Motion intelligence,
engineered for reality.

I'm Jarvis Men (门嘉伟). I build reliable control systems for humanoid robots—connecting task-level intent to smooth, constraint-aware motion on real hardware.

CONTROL / 001250 HZ
01
TASK SPACEDesired motion
IN
constraints
02
WHOLE-BODY SOLVERHQP / IK / MPC
q, q̇, q̈
03
ROBOT RUNTIMESafe execution
LIVE
01

Expertise

Control is where intelligence
meets the physical world.

I work across algorithms, runtime architecture and hardware integration—the layers that determine whether a robot motion is merely possible or genuinely reliable.

A

Whole-Body
Control

Hierarchical objectives and constraints across coupled mobile and humanoid systems.

  • HQP / QP
  • Redundancy resolution
  • Joint & workspace limits
B

Motion
Generation

Continuous trajectory generation from sparse goals and streamed commands.

  • IK / FK / Jacobians
  • VAJ-constrained smoothing
  • Time scaling
C

Execution
Runtime

Deterministic command flow, observation and recovery around real robot motion.

  • ROS / ROS 2
  • Watchdogs & freshness
  • Mode arbitration
D

Robot
Learning

Bridging learned policies with classical constraints and hardware-aware execution.

  • VLA execution
  • Imitation learning
  • Dataset pipelines
02

Selected projects

Selected engineering
case studies.

Representative systems spanning whole-body control, learned-policy execution and production-oriented manipulation.

P/01
2×7DoF arms3DoF torso250 Hz

Humanoid Dual-Arm Whole-Body Control

A real-time whole-body motion stack coordinating two 7-DoF arms and a 3-DoF torso from end-effector goals to stable joint commands.

  • Hierarchical QP task priorities
  • Pinocchio FK / Jacobians
  • Joint-limit & singularity handling
  • Redundancy resolution
  • Command smoothing
  • Watchdog & reachability
Humanoid robot coordinating two 7-DoF arms and a 3-DoF torso
P/02
Action chunkVAJ limitsRuntime safety

VLA Motion Execution Runtime

A constraint-aware execution layer that turns model-generated action chunks into continuous, observable and hardware-safe motion.

  • Chunk transition continuity
  • Velocity / acceleration / jerk bounds
  • Adaptive time scaling
  • Command-vs-readback FK
  • Freshness & epoch guards
  • Safe hold semantics
Robot arm executing smooth VLA action chunks with motion constraints
P/03
Mobile baseTorsoDual arms

Mobile Manipulation Whole-Body Controller

Unified coordination for a mobile base, torso and dual arms—balancing reachability, posture and task tracking across the full kinematic chain.

  • Shared task-space objectives
  • Base / arm coordination
  • Workspace expansion
  • Mode arbitration
  • Teleoperation path
  • Constraint priorities
Mobile manipulator coordinating its base, torso and dual arms
P/04
Vision / VLASuctionTactile

Industrial Unstructured Bag Picking

A perception-to-action pipeline for selecting and picking exposed flexible packages in cluttered, production-oriented scenes.

  • RGB-D candidate generation
  • Visible-target ordering
  • Suction point selection
  • Tactile feedback
  • Failure-aware retries
  • Independent test protocol
Robot using RGB-D vision and suction to pick flexible bags from a cluttered bin
03

Robotics stack

From model math
to motor command.

Fluent across the stack required to turn robotics research into robust systems.

LANGUAGES

C++ · Python · C

ROBOTICS

ROS · ROS 2 · Pinocchio · EtherCAT

CONTROL

HQP · QP · IK · FK · MPC · VAJ

LEARNING

VLA · Imitation Learning · Vision

SYSTEMS

Linux · RTOS · SOEM · Real-time Runtime

04

Patents / IP

Applied ideas,
made defensible.

Published patent applications in robot motion planning and control. Unverified titles are intentionally omitted.

05

About

Portrait of Jarvis MenJARVIS MEN · 2026

ENGINEER · BUILDER · SYSTEMS THINKER

I care about the last mile between an elegant algorithm and a robot people can trust.

Trained in Automation at the University of Science and Technology of China, I now develop motion-control systems for humanoid robotics. My work spans whole-body control, trajectory smoothing, singularity-aware IK, robot runtime safety and cross-team integration.

Previously, I built high-precision multi-axis and wafer-transfer robot controllers, achieving ±10 μm system-level commissioning accuracy, ±5 μm wafer-centering compensation and a 4× reduction in PDO communication jitter—from 200 μs to 50 μs.

17Coordinated DoF
250HzCommand stream
30+System fixes
4Patent submissions
2026 — NOWDaimon Robotics

Motion Control Development Engineer · Humanoid & Manipulation

2025Moore Threads

High-Performance Computing Engineer

2023 — 2025SiCarrier

Motion Control Algorithm Engineer · Wafer Transfer Robotics

2019 — 2023USTC

B.Eng. in Automation · Three merit scholarships

Available for meaningful robotics conversations

Let's build motion
that holds up in reality.

jiawei.men@dmrobot.com↗