Custom-built platform
Krish Agarwal
QUADRUPED
ROBOTICS
A custom quadruped, engineered from scratch — mechanical design, electronics, kinematics and software. Not a kit, not a commercial robot: a complete end-to-end robotics project.
01 · Kinematics
Twelve degrees of freedom.
Four legs, each a three-joint serial chain — hip (coxa), upper leg (femur) and lower leg (tibia). The whole mechanical hierarchy lives in a single URDF that the browser assembles in real time, joint limits and all — the same file that drives simulation.
- Legs
- 4
- Joints / leg
- 3
- Total DOF
- 12
- CAD
- Fusion 360
02 · Actuation
Twelve joints,
one control path.
Each joint is a DS3240MG high-torque digital servo, driven over PWM through a PCA9685 controller. A calibration pipeline and joint-level software abstraction mean every servo is commanded the same way — pose and gait targets, not raw pulses.
- Servos
- 12 × DS3240MG
- Driver
- PCA9685
- Signal
- PWM
- Hip range
- ±45°
03 · Structure
Designed to be iterated.
Every structural component was modelled in Fusion 360 and optimised for 3D printing, modularity and ease of maintenance. The central body carries the compute, IMU and power distribution — a platform built to be taken apart and improved.
- Structure
- 3D-printed
- CAD
- Fusion 360
- Legs
- Modular
- Source
- URDF
04 · Teardown
Exploded.
Scroll to pull BARQ apart, component by component. Hover any part for its engineering spec.
Chassis
Structure
Coxa · Hip
Actuation
Femur
Kinematics
Tibia + Foot
Contact
05 · Technical
The chain,
drawn to scale.
Each leg is a three-link serial manipulator. The foot position is a closed-form function of three joint angles — the same forward kinematics the browser solves in real time.
06 · Robot Lab
Take the controls.
Drag to orbit, pinch or scroll to zoom, then drive the robot live — poses, gaits and demos across seven environments.
07 · Engineering Pipeline
From CAD to hardware.
The path every capability takes — modelled, described, simulated, then proven on the physical robot. Autonomy is the next stage, not a current claim.
- 01
Mechanical CAD
Full robot modelled in Fusion 360.
- 02
URDF Generation
CAD exported to a URDF — the single source of truth.
- 03
Simulation
Kinematics validated in Webots and the web viewer.
- 04
Visualization
Live 3D visualization and debugging tools.
- 05
Electronics
Jetson, PCA9685, power and sensors integrated.
- 06
Calibration
Per-servo zeroing and joint-level abstraction.
- 07
Motion Development
Forward / inverse kinematics and pose control.
- 08
Hardware Validation
Poses and stances verified on the real robot.
- 09
Autonomous Behaviours
PlannedNavigation and perception — planned.
08 · Technical Highlights
Why it’s built the way it is.
Twelve deliberate decisions — from the URDF single-source-of-truth to real hardware validation.
12 DOF Architecture
Three joints per leg give each foot full 3D placement for legged locomotion.
Custom URDF
One description drives simulation, visualization and control — no drift between them.
Jetson Compute
Edge GPU headroom so perception and control live on the robot, not a laptop.
Integrated IMU
Orientation feedback is the foundation for balance and closed-loop motion.
LiDAR Ready
360° ranging is on-board so mapping can be developed without new hardware.
Vision Ready
CSI cameras are wired in for future detection and visual SLAM.
3D-Printed Chassis
Printed parts make every link cheap to iterate and easy to repair.
Fusion 360 Design
Parametric CAD keeps the mechanical design modular and revisable.
ROS Compatible
Standard middleware so the stack can grow into the wider robotics ecosystem.
Simulation Pipeline
Motion is proven in simulation before it ever touches a servo.
Modular Electronics
PCA9685 + isolated power let subsystems be swapped independently.
Real Hardware Validation
Every capability is tested on the physical robot, not just on screen.
09 · Hardware
Every subsystem, integrated.
Compute, actuation, power and sensing — real components wired together on a custom platform. 'Integrated' runs today; 'Ready' is on-board with its software still in development.
NVIDIA Jetson Orin Nano
On-board Linux computer. Chosen so ROS, computer vision and future autonomy can all run on one edge platform instead of a tethered PC.
HW-290
Body-orientation sensing for balance and stabilization — the feedback source for future closed-loop locomotion.
12 × DS3240MG
High-torque digital servos — three per leg. Every joint shares one calibrated control path so poses and gaits are reproducible.
4S LiPo · 6400 mAh
A dedicated high-current buck converter feeds the servo rail so twelve simultaneous actuation loads don't brown out the compute.
PCA9685
16-channel, 12-bit PWM controller over I²C. It off-loads precise pulse timing so the Jetson issues joint targets, not waveforms.
3D-printed · Fusion 360
Custom structural parts modelled in Fusion 360 and printed. Designed for modularity, easy maintenance and rapid iteration.
YDLIDAR G2
360° 2D ranging on-board and wired in — targeted at mapping and environment perception as the navigation stack comes online.
IMX219 CSI
CSI camera input for the vision subsystem — the entry point for future object detection and visual SLAM.
10 · Software Stack
A modular stack.
Python services on ROS drive the control layer; a URDF describes the robot for both simulation and visualization. Each layer is independent, so kinematics, sensing and perception can evolve on their own.
11 · Capabilities & Roadmap
What works today. What comes next.
This platform is a work in progress. Everything on the left runs on the robot now; everything on the right is planned engineering, stated as such.
- Stable standing
- Inverse kinematics
- Body pose control
- Servo calibration
- Real-time visualization
- Modular software architecture
- Sensor integration
- Simulation compatibility
- Dynamic walking
- Improved gait generation
- Closed-loop balance
- Terrain adaptation
- Autonomous navigation
- SLAM
- Computer vision
- Obstacle avoidance
- Mission planning
- AI-assisted locomotion
- Sim-to-real workflow
- Edge AI perception