About
SCOUT is a fully 3D-printed, low cost quadruped robotic dog that I built over the course of 4 month. I built this robot for the German science competition Jugend-Forscht. In 2025 I participated the first time in a robotics competition and with SCOUT I received a 2nd price on the state level.
I set my self the goal to develop an autonomous robotic dog by my self to help visually impaired.
All source files can be found on my GitHub: github.com/jacobutermoehlen/SCOUT_RobotDog
Features
primitive walking
straight
angled
wireless communication
dancing
self-made rubber feet
Specifications
12-DoF
Weigth: 3.5 kg
Body length: 400 mm
Body width 160 mm
Height range: ca. 250 mm
8 Sensors
2 Cameras
Design
My robot dog tries to imitate real life dogs, therefore has 4 individually controlled legs with each 3-DoF. Together the dog has 12-DoF and 12 actuators. The design process started by designing a single test leg.
The legs had to fulfill the following requirements:
strong
fairly lightweight
low inertia
cost effective
For the leg to be strong, yet lightweigth, all structural component are 3d printed from PLA+ with a high wall count. Both limbs have a lattice inner structure for more weight savings. The 3rd actuator connects to it corresponding joint via a parallel linkage, therefore the centre of mass is is located closer to the ratational axis if joint 1 and 2 and the overall inertia is reduced.
The rest of the mechanical components don’t include any actively moving parts and are therefore quite simple. The robot is designed with modularity in mind, every part can be accessed easily and the electronics compartment can be removed without taking anything apart.
CAD assembly of SCOUT
all mechanical parts
Hardware
The brain of SCOUT is a NVIDIA Jetson Orin Nano Dev Kit, which I chose because of the performance. I considered using a much cheaper Raspberry Pi, but the Raspberry Pi lacks power for future implementations of computer vision and reinforcement learning.
SCOUT has multiple sensors to perceive its enviroment. 5 HC-SR04 measure distances left, right, front left, front right, and front down. A VL53LOX laser sensor measures distance in the front. Along these distance sensors SCOUT includes a INA219 sensor for voltage measuremens and a ICM20948 MPU for acceleration measurements. All sensors are connected to an ESP-32, my favorite low-cost microcontroller. Over serial all the sensor data is sent to the NVIDIA Jetson.
Every joint is actuated by 40 kg/cm RC-servos through a PCA9685.
All real-time calculatoions and path planning is done on the NVIDIA Jetson, which then sends corresponding positions to the PCA9685 over I2C.
electronics component
Inverse Kinematics
With the design of the legs finished, I moved on to calculating the inverse kinematics. The inverse kinematics is a key-aspect in robotics. The inverse kinematics consist of a set of equations that calculate each joint angle needed for the leg to reach its desired end-position.
Because each leg of the robot move completely independent, the inverse kinematics only needed to be solved for a 3-DoF mechanism, which wasn’t that challenging.
With the inverse kinematics solved, each leg could now reach its desired end-effector position in a cartesian system.
The solving of the inversed kinematics was a crutial step, needed for the creation of the walking cycle.
Walking
Making SCOUT walk was by far the most challenging part of this project, due to the complexity of the motion system and the overall stability of a quadruped robot.
My initial approach was to make a foot trajectory for each of the four legs. Using two Bezier-Curves I created a foot trajectory.
I copied this trajectory over to every leg, with a phase shift, so that two diagonal legs contact the ground simultaniously.
When supposed to turn, new foot trajectories are calculated for the forelegs.
ground contact - straight
Issues with walking
Using this “hard-coded” method of defining the walking sequenz did’t work well. When trying out the walking, after only a few steps SCOUT would fall over. Over a of 2 weeks I tried out every combination of walking speed, center distance of the legs, walking height, length of ground contact, etc. without any major improvements.
ground contact - curve
foot trajectory
Custom Silicone Feet
While conducting the walking tests of SCOUT I noticed that the feet where slipping and therefore SCOUT wouldn’t move foreward. Until this point the feets where made of cut bouncy balls glued to an adapter. This connection wasn’t secure as the bouncy balls would often just fall of and the feet did’t have enough grip.
To improve the grip, strength and aesthetics I decided to make my own silicone feet. I created a mold in which I poured liquid silicone around a leg stump. After a few hours of curing I could remove the leg and had a perfectly molded silicone feet.
iteration of SCOUT’s feet
Future Perspectives
Eventhough I have already participated with my robot in the Jugend-Forscht competition this project is not over yet. There are many things that need improvement. Currently (May 2025) SCOUT is not able to walk reliably. He only manages to walk a few steps until he falls over. In the future I’ll implement a reinforcement learning algorithm with which SCOUT can learn to walk by himself.
As I have overlooked the disadvantages of quadruped robot dog, mainly stability, I will also look into mounting wheels to SCOUT’s feet and only use the legs to climb obstacles.
When planning this project I envisioned a fully autonomously robotic dog that can help visually impaired people navigate more easily. Eventhough I therefore included all the necessary hardware by default and future hardware upgrades will be unlikely. To make SCOUT autonomous I will utilize the computing power of the Jetson Nano to perform computer vision and eventually a small neural network.
SCOUT is still a work in progress, with the finish line in sight, but not yet crossed.