Nature has long been a source of engineering inspiration. In the realm of robotics, bio-inspired design offers an avenue to develop efficient, resilient, and adaptive machines. One such endeavor is CERBRUS — a quadruped robot designed to mimic the anatomy and function of ants. Developed as part of my B.Tech thesis at IIT Hyderabad, CERBRUS explores how merging locomotion and manipulation into a single head-mounted system can redefine robotic capability.
1. The Problem Statement
Conventional quadruped robots with back- or torso-mounted manipulators face several limitations, including:
• Increased complexity in kinematics and control
• Shifting center of mass that can destabilize locomotion
• Reduced reach and interaction efficiency in narrow or constrained environments.
There is a need for a lightweight, efficient manipulation system that can be integrated with a quadruped platform without compromising its mobility or balance. Nature’s solution—seen in ants and similar organisms—suggests that a head-mounted gripper can be both compact and functional.
 |
| Ant mouth like Gripper Design |
2. Mechanical Aspect
The first step in any successful product is the mechanical design and analysis. For this project I have used SolidWorks (ver 2023) for the computer designs. The following are some of the milestones in design:
• Initial version had a small shoulder size as well as bulky legs. Approximate weight was 1.7kg.
• Next iteration was done to change the shoulder size and was almost increased by one point five times.
• This increase help increase the area of the contact polygon, meaning the robot became more stable.
• Further iterations all focused on the leg design as well as making the legs as light as possible to reduce the inertia while swinging.
 |
| Isometric view of the final CAD assembly without gripper |
3. Electronic Aspect
The electronic architecture of CERBRUS was meticulously designed to support real-time processing, precise control, and seamless communication between modules. The onboard computing unit consists of a Raspberry Pi 5 (8GB) running Ubuntu 24.04 and ROS2, supported by two Arduino Mega boards and one Arduino Nano. One Arduino Mega board is exclusively for driving servos whereas other Arduino Mega is for interfacing servos. The Arduino Nano is used for telemetry.
Twelve Waveshare ST3215 servo motors drive the leg joints, while two SG-90 servos actuate the gripper mechanism. Power is delivered via a 3S 5200mAh battery through dual 5V 5A DC-DC buck converters. To ensure orientation and stability, a BNO055 IMU sensor is deployed. Long-range communication is achieved using a pair of HC-12 RF modules, while a Logitech web camera facilitates visual input for computer vision tasks.
The power delivery system is optimized for the high current demands of the Raspberry Pi 5, which requires a steady 5V 5A supply. A buck converter is employed to step down voltage from the 12V battery, with a USB-C connector soldered directly for secure Pi connectivity. Communication between the robot and controller is handled via HC-12 modules operating at frequencies ranging from 400 MHz to 500 MHz, ensuring reliable long-range transmission. Additional safety and utility features include voltage sensors and HDMI/USB-C extenders for debugging and developer access.
 |
| Onboard Quadruped Circuit |
4. Software Aspect
The software stack of CERBRUS is designed for modularity, real-time control, and ease of expansion. It operates on Raspberry Pi 5 embedded within the quadruped running Ubuntu 24.04 with ROS2 Jazzy.
The communication between the Raspberry Pi and Arduino boards is through the USB port on the pi. The baud rate for data transfer for the Arduino Mega Servo Driver is 1000000 and for the Arduino Mega sensor suite is 9600. The control loop operates at 103 Hz. The motion command from the raspberry pi is given as a byte stream of height, y, z and velocity of each leg along with potentiometer values that are mapped to gripper servo angles. So, overall, 18 values, plus two safety check bytes (-1) are sent.
 |
| System Overview |
5. Future Work
While the current prototype validates the concept effectively, several enhancements are proposed for future development:
• Sensor Integration: Incorporating Lidar, force sensors, and better vision systems to improve balance, perception, and autonomy.
• Terrain Adaptation: Implementing terrain classification and adaptive gait strategies for traversal over irregular surfaces.
• Gripper Enhancement: Upgrading the gripper for dynamic control, precision gripping, and feedback-based manipulation.
• Machine Learning: Applying reinforcement learning or imitation learning for self-optimization of gait and gripping behavior.
• Energy Efficiency: Exploring power management strategies and lightweight materials for extended operation.
• Swarm Capabilities: Developing swarm behaviors for cooperative transport or environ mental interaction.
6. References
The demonstration of the completed project can be found in the YouTube video linked below:
You can also contribute to this exciting project by checking this Cerbrus Repository:
I will add more blogs to cover the basic concepts of quadrupeds, math and algo behind it, as well as some project logs and guide hacks. So, stay tuned!
Comments
Post a Comment