Independent Research · Leesburg, Virginia

Engineering theFuture of Walking

ExoLab is a solo R&D lab run by Greg Klassen — a 30-year DSP engineer, retired CEO, and SCA3 patient — dedicated to building open exoskeleton control systems for neurological mobility conditions.

Real hardware. Real code. Real stakes.

PHASE 4 TARGET
🦾
PHASE 4 TARGET
Wearable · CubeMars AK80-9 Actuated
// render coming soon
↗ Goal: Wearable single-joint knee assist · CubeMars AK80-9 actuated
0 Years DSP Experience
0 Build Phases Planned
0 Phase 1–3 Budget (K)
0 Target Joint Torque

A DSP Engineer Who Needs an Exoskeleton

Greg Klassen spent three decades in digital signal processing — designing real-time filter architectures, tuning PID loops, and building embedded control systems from the ground up. He then spent years as a CEO, before a diagnosis of Spinocerebellar Ataxia Type 3 (SCA3) changed the plan.

SCA3 is a progressive neurological condition that attacks coordination, balance, and gait. Commercial exoskeletons exist — but none are designed for cerebellar ataxia, where the challenge is balance, not paralysis.

So Greg is building his own. Working from a six-acre property in Leesburg, Virginia, ExoLab is a one-man R&D operation applying 30 years of signal processing expertise to the hardest problem Greg has ever faced: getting himself back on his feet.

The lab pursues three goals simultaneously: personal mobility research, rigorous open documentation of the build process, and informed engagement with the commercial and investment exoskeleton landscape.

Current status: Phase 1 hardware on order. CubeMars AK80-9 actuators inbound. Software stack under development.

// DSP → Exo Control Mapping
PID Loop Tuning
Real-Time Filtering
State Machine Design
Embedded C / STM32
System Identification
Sensor Fusion / IMU
CAN Bus Protocol
Impedance Control

The ExoLab Build Plan

A five-phase roadmap from benchtop motor control to a wearable bilateral exoskeleton assist system. Built on 30 years of DSP experience. Click any phase for details.

PHASE 01
Benchtop Motor Control
~$800
PHASE 02
Gait Sensor Platform
~$225
PHASE 03
Simulated Knee Joint
~$220
PHASE 04
Wearable Single-Joint
~$2,900
PHASE 05
Bilateral Hip + Knee
~$15,000+

PHASE 01 — Benchtop Motor Control

One motor. One controller. Your own PID loop. The CubeMars AK80-9 delivers 9Nm rated / 22Nm peak torque through an integrated brushless motor, planetary gearbox, encoder, and CAN driver — all in a single package. The MIT Mini Cheetah protocol runs over CAN bus. Objective: command torque, read position, tune a real impedance controller.

ComponentPartCostNotes
ActuatorCubeMars AK80-9 V3.0~$50048V, MIT CAN protocol, integrated encoder
CAN InterfaceWaveshare USB-CAN-A~$22socketCAN on Linux, plug-and-play
Power SupplyMEAN WELL 48V 10A~$90Bench PSU — current-limited for safety
E-Stop22mm latching mushroom switch~$12Non-negotiable. Within arm's reach always.
Frame2020 Aluminum Extrusion~$50Rigid mount — 9Nm will move a light table
ControllerRaspberry Pi 5 or STM32G4~$60Python CAN stack or embedded C

PHASE 02 — Gait Sensor Platform

Pure sensing. No actuation. Strap IMUs and force-sensitive resistors to both legs and walk — on carpet, on gravel, on the six-acre property. Capture what SCA3 gait actually looks like at 500Hz. This data becomes the training signal for every control algorithm written in Phases 3–5.

ComponentPartCostNotes
IMU (×3)ICM-42688-P breakout board~$30 ea32kHz sampling, significant upgrade from MPU-6050
Force SensorsInterlink 402 FSR (×8)~$6 eaHeel, 1st/5th met head, big toe per foot
ADCADS1115 16-bit, I2C~$134-channel analog read for FSRs
MCUTeensy 4.1 (600MHz ARM M7)~$30Real-time capable; DSP-friendly architecture
StorageMicroSD module + 8GB card~$8Log raw data at 500Hz for offline analysis
BatteryLiPo 3.7V 2000mAh + TP4056~$15Wearable power for walking trials

PHASE 03 — Simulated Knee Joint

Connect the motor to a 1:1 mechanical replica of a human knee. Drive it with gait data captured in Phase 2. When this mechanical leg tracks a walking cycle correctly — under impedance control — you are ready for Phase 4. Control loop latency target: under 5ms end-to-end.

ComponentPartCostNotes
Frame80/20 1" aluminum extrusion~$100Vertical leg mock-up with pivot joint
Knee PivotFlanged pillow block bearing 12mm~$25Zero-backlash coupling to motor shaft
Load Cell50kg S-type + HX711 amp~$22Measure actual torque output, validate commands
Current MonitorINA226 I2C power monitor~$10Real-time power consumption → battery sizing
Leg SegmentsAluminum flat bar 1"×1/4"~$20Thigh + shank fabricated from hardware store stock

PHASE 04 — Wearable Single-Joint Assist

The first device that goes on a body. A powered knee orthosis using the AK80-9 to provide assistive torque — not walking for you, but assisting where your own muscles are already trying. Safety gate: Phases 1–3 must be complete with 4+ weeks of stable operation before this phase begins. Katia present for all wearable sessions. Dr. Rosenthal signed off.

ComponentPartCostNotes
Second ActuatorCubeMars AK80-9 (left leg)~$500Bilateral is safer than unilateral for ataxia
Battery48V 20Ah wearable LiPo pack~$250Sized for 2hr walking sessions
Brace FrameKAFO modified or custom carbon~$1,000Structural foundation for the actuator mount
RF E-StopWireless kill switch handheld~$75In-hand at all times during walking trials
Safety HarnessChest harness + overhead tether~$100Mandatory for first 20+ wearable sessions

PHASE 05 — Bilateral Hip + Knee System

The full system: hip and knee actuation, left and right, with a unified control stack running gait phase detection from dual IMU/FSR sensor arrays. Four AK80-9 actuators. A full power electronics bay. This is what academic labs build with grad student teams. ExoLab will build it solo — which means it will take longer, and every decision will be deeply understood. Estimated timeline: 6–18 months beyond Phase 4.

ComponentEst. CostNotes
2× Additional AK80-9 (hips)~$1,000Higher torque variant for hip flexion/extension
Power Electronics Bay~$40048V BMS, current distribution, emergency cutoff
Bilateral Frame~$2,000Carbon fiber + aluminum hybrid structural frame
Central Controller~$200Jetson Nano or Teensy 4.1 — real-time gait FSM
Full Sensor Array~$3006× IMU, 16× FSR, full bilateral coverage
Misc / Integration~$1,0003D printing, machining, wiring, iteration

⚡ DOWNLOAD FULL LAB PLAN — The complete ExoLab Build Plan document including detailed parts lists, control theory background, DSP mapping tables, workshop setup guide, and safety protocols is available as a Word document. Contact to request →

30 Years of DSP → Exoskeleton Control

The translation from digital signal processing to exoskeleton control is nearly one-to-one. Every skill developed over a 30-year engineering career maps directly to the core challenges of robotic mobility.

DSP Background
Real-time filter design
ExoLab Application
Kalman / Madgwick filter on IMU data for gait phase detection
DSP Background
PID loop tuning
ExoLab Application
Torque / impedance control on the AK80-9 actuator — same math, mechanical plant
DSP Background
Noise floor & SNR analysis
ExoLab Application
Heel-strike detection from force sensors buried in noisy floor contact signal
DSP Background
Fixed-point arithmetic & embedded C
ExoLab Application
Real-time control loop on STM32G4 — sub-5ms latency target
DSP Background
State machine design
ExoLab Application
Gait cycle FSM: heel-strike → loading → midstance → toe-off → swing
DSP Background
System identification
ExoLab Application
Characterizing the mechanical plant: joint stiffness, inertia, damping coefficients
DSP Background
Fourier analysis & spectral methods
ExoLab Application
SCA3 gait variability analysis — trunk sway frequency spectrum quantification
DSP Background
CEO / product leadership
ExoLab Application
Investment analysis of exoskeleton companies (Lifeward/LFWD, Wandercraft, Ekso)

From Bench to Walking

A realistic timeline. Each phase is independently satisfying and produces real, documented results. No phase is skipped.

NOW · Spring 2026
Hardware Ordered. Software Stack Under Development.
CubeMars AK80-9 actuators ordered. Waveshare USB-CAN-A interface acquired. Python-CAN installed on Chromebook Linux (ARM64/Crostini). MIT Mini Cheetah protocol under study.
ACTIVE
PHASE 1 · Summer 2026
First Torque Command. First PID Loop. First Data.
Motor spinning under own control code. Position, velocity, and torque logged. Impedance controller tuned. E-stop tested. Lab bench established at Canby Road property.
~$800
PHASE 2 · Fall 2026
First Gait Dataset. SCA3 Signature Characterized.
IMU + FSR sensor rig worn during walking trials on property and W&OD Trail. Gait phase FSM validated. SCA3 trunk sway quantified. Dataset potentially shared with NAF research community.
~$225
PHASE 3 · Winter 2026–27
Mechanical Knee Tracks Real Gait. Control Latency Validated.
Simulated knee joint reproduces captured gait cycles under impedance control. Control loop latency measured and confirmed under 5ms. Load cell validates torque commands.
~$220
PHASE 4 · 2027
First Wearable Session. With Katia. With a Tether.
Bilateral knee assist device worn for the first time. RF e-stop in hand. Overhead safety tether. Dr. Rosenthal consulted. Every session documented. This is the milestone everything else points toward.
~$2,900
PHASE 5 · 2027–2028
Full Bilateral Hip + Knee System.
Four actuators. Full gait control stack. Walking outdoors on six acres in Leesburg, Virginia. Independent. On your own terms.
~$15,000

The ExoLab Reading Stack

The foundational texts and repositories behind this build. Every resource here is free and directly applicable to the control problems ExoLab is solving.

Live from the Field

Greg attended the Robotics Summit & Expo in Boston — the world's leading technical event for commercial robotics developers, produced by The Robot Report and WTWH Media. Tracking exoskeleton hardware, rehabilitation robotics, and commercial mobility devices.

// Greg's Session Notes · via Otter.ai
NOTES COMING SOON
Session transcripts from the floor will be posted here. Greg is attending May 27–28 with a focus on exoskeleton control systems and Medicare pathway sessions.

// Raw session notes — unfiltered, from the floor. Not affiliated with WTWH Media.

// Conference Resources
ROBOTICS SUMMIT & EXPO 2026
Official Conference Site
Boston Convention & Exhibition Center · May 27–28 · 6,000+ developers · 250+ exhibitors · 50+ sessions
GREG'S FOCUS AREAS AT THE SUMMIT
  • → Exoskeleton control systems & CubeMars actuator hardware
  • → Medicare/insurance pathways for wearable mobility devices
  • → ReWalk 7 · Ekso Indego · Wandercraft Atalante X comparisons
  • → Intent detection & ground truth methodologies for SCA3
  • Conor Walsh (Harvard SEAS) — wearable robot session
  • → IROS 2026 Pittsburgh preview & networking

Collaborate with ExoLab

ExoLab welcomes connection from researchers, clinicians, engineers, investors, and anyone navigating a neurological mobility condition. This work is open and documented.

Founder
Greg Klassen — DSP Engineer, Retired CEO
📍
Location
Leesburg, Virginia (Loudoun County)
🌐
Domain
exolab.llc
🔬
Research Interest
SCA3 · Exoskeleton Control · DSP · Gait Analysis
📻
Also Publishing
"What's Next?" Podcast — Post-career identity & purpose

// This is an active research lab, not a product company. Messages are read personally by Greg Klassen. Response time: 24–48 hours.