An AI-first Mars rover that thinks before it traverses
BRACU Mongol-Tori's 2022 entry, "Mongol-Tori Encephalon," is a next-generation Mars rover built to assist astronauts and conduct scientific research at URC 2022. This iteration is defined by an AI-emphasized redesign across the autonomous, network, control and mechanical systems, with the science and electrical systems rebuilt entirely from scratch. A four-wheel modified rocker-bogie platform carries a six-DOF arm, dual-band vision, and an onboard science laboratory for atmospheric and soil analysis.
This year's headline is an Artificial-Intelligence-emphasized overhaul branded 'Encephalon,' adding autonomy across tool classification, equipment detection, navigation, and science with the electrical and science systems rebuilt from scratch.
A custom-made double-sided linear actuator with a custom planetary gearbox replaces the old extender, cutting rover weight while widening the base to prevent flipping during vertical drops.
Two power-independent communication systems pair a peer-to-peer 2.4 GHz control/IP-camera link for >1 km range with a 5.8 GHz analog vision system to eliminate long-range lag.
With the RTK GPS unavailable due to the pandemic, the team built a custom differential-GPS algorithm on standard modules, fused with MPU9250 IMU and RPLiDar A1 mapping on a Jetson Nano for accurate waypoint navigation.
The science module adds YOLOv5 microscopic rock classification and an LSTM weather-forecasting model trained on Curiosity Rover REMS data, alongside a pyramid soil box running biomass, capillary, amino-acid and starch chemical tests.
Every discipline on the team owns a slice of the machine. Here is how each one comes together.
A modified four-wheeler platform with an H-shaped ladder space-frame chassis and a modified rocker-bogie suspension. Major updates this year focused on a custom double-sided linear actuator extender, a redesigned wheel, and an upgraded six-DOF arm and three-finger claw.
Two independent communication sub-systems on 2.4 GHz and 5.8 GHz, each with detached power distribution. A peer-to-peer 2.4 GHz link handles control and IP-camera data, while a 5.8 GHz analog vision system overcomes long-range lag.
The electronics subsystem was redesigned entirely, split into purpose-specific boards: power management, divider, manual/debug, and controller boards. Power flows from the Li-Ion battery through a kill-switch and MOSFET protection.
The rover runs ROS Melodic on Ubuntu 18.04 with the master node on a portable base station. A publisher/subscriber model with an MQTT bridge handles commands, and PyQt5 GUIs drive control, arm IK and science data analysis.
The Autonomous Navigation System runs on an NVIDIA Jetson Nano linked to an Arduino Mega over serial. It fuses IMU, GPS, and LiDAR for waypoint navigation, with OpenCV/YOLO-based detection of AR tags and street signs.
A from-scratch science module split into atmospheric and soil analysis to search for past or future signs of life. An onboard sensor box acts as a laboratory, and a pyramid-shaped soil box runs biomass, capillary, amino-acid and starch tests, now augmented with AI rock classification and weather forecasting.
How this rover is engineered to score across every University Rover Challenge task.
The modified rocker-bogie suspension, strong chassis and jute-rope wheels let the rover traverse up to 1 km of rocky, sandy terrain while absorbing jolts and staying stable on vertical drops. The durable arm lifts heavy toolboxes and equipment, with a TensorFlow object-detection AI (CNN + SVM regression on bounding boxes) classifying tools and an OpenCV + Sharp distance sensor estimating range.
A new three-finger claw with a feedback end effector provides precise, IK-measured manipulation for switches, knobs, latches and cache boxes. Light AI models on the Jetson Nano detect elements like toggle switches, joysticks, keyboards and first-aid boxes to instruct the driver.
Jetson Nano + Arduino Mega drive autonomous waypoint navigation using IMU heading, custom DGPS localization and RPLiDar A1 mapping/path planning. The rover scans plus/minus 90 degrees for AR tags (OpenCV ARUCO 4x4_50), estimates distance via a calibrated polynomial, and identifies street signs with YOLO v5.
A two-part module performs atmospheric analysis via an onboard sensor-box laboratory and soil analysis via probes and a pyramid test box, testing for biomass, water capillary action, amino acids and starch to assess past and future life and environmental change on Mars.
2026Lighter, cooler, CAN-smart: a rocker-bogie rover engineered for Mars terrain
2025Redundant by design, autonomous by ambition — Hypersonic conquers Mars terrain
2024Reborn for Mars: omni-wheel agility, a 7 kg-payload arm, triple-layer comms
2023AI-powered Mars rover with an autonomous onboard planet analyser
2020Centralized power, autonomous traversal, and an onboard Mars-soil laboratory
2019Fourth-generation Mars rover, rebuilt for retrieval, servicing, autonomy and science