WIM Robot Control Platform — NVIDIA Partnership

Executive Summary

Why WIM for NVIDIA Jetson?

Certifications — All Passed

4× KOTCA performance certifications + 7× KIRIA robot evaluations (payload, repeatability, speed, picking). Plus ISO 9001/14001/45001.

Real-World Deployments

Waste sorting, crop harvesting, parts machining, process automation — with paying customers and verifiable performance.

24×

Jitter Reduction on Jetson Orin

358µs → 15µs worst-case jitter via PREEMPT_RT + CPU isolation on NVIDIA Jetson Orin. Industrial-grade determinism at the edge.

21.2%

Motor Control Accuracy Gain

Acceleration feedforward (LPF α=0.3) yields 21.2% position accuracy improvement with RL-based friction compensation — certified in real robot deployments.

99%

Vision Accuracy, Certified

PP/PE waste sorting at 99.1% mAP@0.5, 142fps TensorRT FP16 on Jetson — KOTCA certified. TensorRT + DeepStream pipeline fully optimized for Jetson edge AI.

Pure Software Stack

WIMPACK runs on NVIDIA Jetson Orin. No proprietary hardware lock-in. Compatible with industrial robots, humanoids, mobile platforms.

Layer	Component	Key Spec
Application	ROS2 Application	Task orchestration
AI Vision	TensorRT + DeepStream	142fps FP16, 99.1% mAP
Motion	Trajectory + Dynamics	1kHz servo loop
Real-Time	PREEMPT_RT + Lock-free IPC	<1µs IPC latency
OS	PREEMPT_RT Jetson	15µs jitter
HW	NVIDIA Jetson Orin	275 TOPS

Technical Differentiation

What Makes WIMPACK Different

Friction Compensation

RL-based friction model identification + real-time compensation on Jetson Orin. Certified performance in actual robot experiments.

Metric	Before	After	Improvement
Position RMSE (Combined B3)	0.001389 rad	0.000849 rad	−38.9%
Position Error RMS	High	−39%	Certified
Torque Ripple	Visible	Eliminated	Near-zero
Stall at low speed	Yes	No	Full range

Acceleration Feedforward Tuning

Systematic feedforward gain optimization — best position accuracy with stable torque output.

Method	Position RMSE (rad)	Torque RMSE (Nm)	Notes
Baseline (no FF)	0.001389	2.77	Reference
Numerical diff	0.001141 (−17.8%)	4.29 (+55%)	Torque unstable
LPF α=0.3 (WIM)	0.001095 (−21.2%)	2.68 (−3.2%)	Best trade-off

Final benchmark: WIM acceleration feedforward vs. baseline methods

Position error (Joint 1) — before and after friction compensation

PREEMPT_RT kernel on NVIDIA Jetson Orin with CPU isolation. Achieves industrial-grade determinism required for 1kHz servo control.

Before (Standard Kernel)

After (PREEMPT_RT + CPU Isolation)

Configuration	Max Jitter	P99 Latency
Standard kernel	358µs	~120µs
PREEMPT_RT only	80µs	35µs
+ CPU Isolation (WIM)	15µs	8µs

WIMPACK RT Certification (KST-25-236)
Parameter	Certified Value
Control Cycle	0.0232 ms
Cycle Jitter	0.0042 ms
Standard	KOTCA KST-25-236

Note: 15µs = worst-case max latency under stress. Certified avg cycle jitter = 4.2µs (KOTCA KST-25-236)

Latency timeseries: standard vs. PREEMPT_RT vs. CPU-isolated

Max latency across CPU isolation configurations

PlotJuggler-based real-time monitoring system — live visualization of RT control metrics

Full TensorRT + DeepStream pipeline on Jetson. KOTCA-certified vision performance for production waste sorting robots.

Pipeline	FPS	Accuracy	Platform
PyTorch FP32 (baseline)	67fps	baseline (mAP 99.1%)	GPU server
TensorRT FP16 (WIM)	142fps	−0.1% mAP (99.0%)	Jetson Orin AGX

Certification	Result
PP Detection AP	99%
PE Detection AP	99%
Throughput (certified)	52fps (W-Ecobot initial model)
Throughput (TensorRT FP16)	142fps (2.1× — post-optimization)
Standard	KOTCA KST-23-034

TensorRT optimization workflow on Jetson Orin

Inference pipeline latency comparison

Kabsch algorithm-based robot calibration with SVD — precise camera-to-robot coordinate transformation without manual tuning.

Method	Error (mm)	Setup Time	Repeatability
Manual calibration	±8mm	4h	Poor
DLT homography	±3mm	30min	Medium
Kabsch SVD (WIM)	±0.5mm	5min	Excellent

Key Properties

Rotation matrix always orthogonal (SVD guarantee)
Works with N ≥ 3 point correspondences
Optimal in least-squares sense
No iterative solver — closed-form, deterministic

Kabsch/SVD coordinate transformation — camera frame → robot base frame

Lock-free SPSC ring buffer for IPC between RT control thread and AI inference thread on Jetson Orin AGX. Eliminates mutex contention and priority inversion.

IPC Method	Latency	Jitter Added	RT-Safe	CPU Usage
POSIX Mutex	2–80µs	High	No (priority inversion)	Medium
ROS2 Topic	100–500µs	Very High	No	High
Shared Mem + Mutex	3–30µs	Medium	Partial	Low
WIM Lock-Free (SPSC)	0.74–0.82µs	Negligible	Yes	Minimal

Implementation Details (ARM64 / Jetson Orin)

Sequential-Consistency ordering for ARM64 weak memory model. Cache-line aligned ring buffer (64-byte). SPSC (Single Producer Single Consumer) design for RT thread safety. Verified with memory ordering analysis on Cortex-A78AE.

Certifications

11 Official Certifications

All performance certifications obtained from Korean government-authorized testing bodies (KOTCA, KIRIA).

KOTCA

KST-25-236

W-RC GPU Controller

Cycle: 0.0232ms · Jitter: 0.0042ms · Real-time control loop

KOTCA

KST-23-034

W-Ecobot Vision System

PP/PE: 99% AP · 52fps · Waste classification on Jetson

KOTCA

KST-23-136

W-Board Embedded Controller

25µs latency · 25W power · 40fps vision

KOTCA

KST-24-156

Vision Detection Model

mAP: 60% · Production vision model for robot guidance

KIRIA ×4

W-Ecobot Performance

Waste Sorting Robot

Payload · Repeatability · Speed · Picking rate — all 4 axes certified

KIRIA ×3

W-Module Performance

Robot Module Evaluation

3× independent performance evaluations by KIRIA Korea

ISO

Quality Management

ISO 9001 · 14001 · 45001

Quality, Environmental, and Safety management systems certified

NVIDIA Ecosystem

Deep Integration with NVIDIA Stack

Isaac Sim / Isaac Lab Usage

Use Case	Tool	Status
Closed-loop articulation test	Isaac Sim	In use (RG6 gripper)
RL policy training (motor)	Isaac Lab	In use
Sim-to-real transfer	Isaac Lab	Validated
Digital twin visualization	Isaac Sim	Deployed
Multi-robot coordination sim	Isaac Sim	In use

Jetson Platform Usage

Jetson Module	Usage	Product
Jetson Orin AGX	RT Control + AI	W-RC, W-Board
TensorRT	Vision inference	All vision products
DeepStream	Video pipeline	W-Ecobot
CUDA	ML training assist	WIMPACK

StereoLabs ZED on PREEMPT_RT

ZED stereo cameras cannot natively run on PREEMPT_RT kernels. WIM collaborated directly with StereoLabs to resolve kernel-level compatibility issues, enabling ZED depth cameras to operate on real-time Jetson systems — a capability previously unavailable in the Jetson ecosystem.

NVIDIA Program Membership

NVIDIA Inception

NVIDIA N&UP — Joint NVIDIA × Korean Gov't startup acceleration program

K-Humanoid Alliance

Customer Reach

Customer & Market Reach

Deployed across 8 industry verticals with 20+ business partners & customers

WIM customer ecosystem across 8 industry verticals

Manufacturing

LG Electronics, SEMES (Samsung), SAMICK THK, HYUNDAI, DN Automotive, Micron, KITECH, SEA, easywarm

R&D & Defense

KIRO (Korea Institute of Robot & Convergence), YASKAWA, Confidential Defense Research Institution

Agriculture & Environment

Gyeongsang Province Agricultural Tech Center, KIGAM, Daehan Industry, J Solution, Kyungpook Nat'l University

Logistics & Autonomous

SAYGO Express, ROGISTICS

Humanoid & Quadruped

NAU Robotics (Prospective)

* All companies listed have been engaged in direct business meetings with WIM Inc.

Competitive Positioning

WIM vs. NexCOBOT

Both companies build on NVIDIA Jetson Orin AGX. Both use EtherCAT. The difference is the layer they focus on.

Feature	NexCOBOT	WIM
Core Business	Controller hardware + middleware platform (OEM B2B)	Application-ready robot control software
Hardware	Designs & manufactures controller boards (GRC, RCB series) with NEXCOM manufacturing infrastructure	W-RC controller (Jetson Orin AGX based). Hardware is the delivery vehicle for WIMPACK software
NVIDIA Relationship	Official Jetson Thor partner, NVIDIA Halos AI Inspection Lab participant, 10+ year partnership	Inception member, N&UP (NVIDIA × Korean Gov't startup acceleration), K-Humanoid Alliance
Safety Certification	TÜV SIL3/PLe, ISO 10218 — industry-leading safety credentials	KOTCA V&V certification, KIRIA robot performance certification. No functional safety cert yet
EtherCAT	10+ years expertise. NexECM master (250µs min cycle), 64 slave support	EtherCAT + CANopen communication. Focused on control algorithms above the fieldbus layer
Control Algorithm	CiA 402 standard motion control. Advanced algorithms not publicly documented	ADRC + RL-trained motor controller at 1kHz on Jetson GPU. Published on tech blog
Application Layer	Not provided — GM stated customers build their own applications on NexCOBOT platform	Directly develops and deploys end-to-end solutions (waste sorting, agriculture, process automation)
Customers	OEM robot builders. 6 published case studies (component supply)	20+ business partners across 8 verticals (LG, SEMES, Hyundai, YASKAWA, etc.)
Company Scale	NEXCOM subsidiary ($160M+ group revenue), 50-100 employees, global offices (5 countries)	Startup (founded 2021), Pre-A funded, Daegu + Seoul offices

How they complement: NexCOBOT excels at the hardware and middleware layer — robust controller boards, EtherCAT mastery, and industry-leading safety certifications. WIM focuses on the software intelligence layer above — adaptive control algorithms (ADRC + RL) and application-level deployment experience. NexCOBOT's GM has stated their goal is to let customers "focus on building innovative applications." WIM is one of those companies actually building those applications on Jetson.

Company

WIM Corporation

Founded	July 2021
Headquarters	Daegu, South Korea
Focus	Physical AI Robot Control Solutions
Products	W-RC (HW) + WIMPACK (SW)
Certifications	11× (KOTCA + KIRIA + ISO)
Deployments	8+ live sites
Website	wimcorp.co.kr
Tech Blog	tech.wimcorp.dev

Key Milestones

2021.07

WIM Corporation Founded

Started with robot control firmware for industrial arms

2023

First KOTCA Certification

KST-23-034: W-Ecobot Vision at 99% AP, 52fps on Jetson

2023

NVIDIA Inception + Pre-A Funding

Joined NVIDIA Inception program. First institutional investment round.

2024

7× KIRIA Robot Certifications

W-Ecobot + W-Module certified across payload, speed, repeatability, picking

2025.01

KST-25-236 RT Control Cert

W-RC: 0.0232ms cycle, 0.0042ms jitter — KOTCA

2025

K-Humanoid Alliance

Selected as member of Korea's national humanoid robot alliance

2026

NVIDIA N&UP Program

Selected for joint NVIDIA × Korean Government startup acceleration program

WIM Robot
Control Platform

Why WIM for NVIDIA Jetson?

WIMPACK — All-in-One SW Stack on Jetson

WIMPACK Software Layers