This research explores a new method for improving the accuracy of RF (radio-frequency) safety simulations used in cellular network design. Traditional ray-tracing models often overestimate safety distances, leading to inefficient use of space and reduced network performance. The study introduces an alternative approach using spherical wave functions to model antenna radiation patterns more accurately. By combining spherical wave modelling with an automated optimisation process, the method produces results that closely match full-physics simulations while reducing errors significantly—by up to 48% in some cases. It also eliminates the need for manual tuning, making the process more efficient for engineers. Overall, the research demonstrates a practical and more precise way to model RF exposure, with the potential to improve network deployment, optimise infrastructure use, and maintain compliance with safety standards.

This blog explores how a Stellenbosch University student developed a cost-effective approach to bipedal robot walking using a robot called Q-Bert. Instead of relying on expensive, advanced platforms, the research focused on proving that reliable walking can be achieved with simpler hardware and smart control systems. By designing an optimised walking trajectory and a dual-layer control system, Q-Bert was able to walk successfully in lab tests, achieving high accuracy and stable movement. The robot completed multiple steps and demonstrated that its walking pattern can be extended for continuous motion. The project highlights that innovative algorithms and control strategies can overcome hardware limitations, making advanced robotics research more accessible. It also lays the groundwork for future improvements, including greater autonomy, better stability, and more complex movements like running.

CubeSpace’s rapid growth as a leading South African space-tech company is closely tied to the Department of Electrical & Electronic Engineering at Stellenbosch University. The company originated from the university’s Electronic Systems Laboratory (ESL), which has a long history of developing real-world satellite technology, starting with Africa’s first microsatellite, SunSAT. The department has built a strong pipeline between academic research and industry, with postgraduate work directly contributing to CubeSpace’s commercial products, particularly in satellite control systems. Key academics and former students have played a major role in transferring knowledge from research into successful space hardware. This collaboration has also helped grow a broader space-tech ecosystem in the Western Cape, supporting local suppliers and innovation. Looking ahead, both CubeSpace and the university aim to expand their capabilities, particularly in AI-driven satellite systems, positioning Stellenbosch as a central hub for South Africa’s emerging space economy.

Professor Bernard Bekker of Stellenbosch University has been appointed as the SARChI NRF Research Chair in Power Systems Simulation, recognising his significant contributions to electrical engineering and renewable energy research. The Chair will focus on advancing power system simulation methods to better understand and plan for the evolving energy landscape, particularly as variable renewable energy and changing consumer behaviour reshape traditional power systems. This work builds on the expertise of the University’s Power Systems Research Group and aims to strengthen South Africa’s national capability in this critical field. With ongoing challenges like load shedding highlighting the need for accurate long-term planning, Professor Bekker’s research will play an important role in improving how power systems are designed, managed, and future-proofed.

This research focuses on developing a low-cost, portable biosensor to detect Fusarium oxysporum f. sp. cubense (Foc), the fungal pathogen responsible for Fusarium wilt—a major threat to global banana production. Traditional detection methods are either unreliable (visual inspection) or expensive and inaccessible (PCR-based testing), making early diagnosis difficult, especially in resource-limited regions. To address this, the study proposes an electrochemical DNA-based biosensor that uses magnetic nanoparticles and screen-printed electrodes to detect the presence of the pathogen quickly and accurately. The biosensor is designed to be cost-effective, portable, highly sensitive, and capable of delivering rapid results in field conditions. By enabling early detection, it can help farmers make informed decisions, improve disease management, and reduce the spread of Fusarium wilt. Overall, the research highlights a promising innovation that could significantly improve plant disease detection and support the long-term sustainability of the banana industry, with potential applications for other plant pathogens in the future.

This research explores how reinforcement learning can be used to enable autonomous vehicles to race at high speeds on tracks they have never seen before. Unlike traditional control methods that rely on pre-mapped tracks, the study introduces a deep reinforcement learning approach (CO-TD3) that uses real-time LiDAR data to guide driving decisions. A key innovation is a “centring term” that helps the vehicle maintain stability and avoid erratic steering, improving both consistency and performance. In simulation, the model outperformed classical and other learning-based methods, achieving faster lap times and a 100% completion rate on unseen tracks. It also demonstrated strong adaptability by avoiding obstacles without explicit training. When transferred to a real-world vehicle, the system maintained near-identical performance to simulation and successfully generalised to new physical tracks without additional training. The study shows that reinforcement learning can match or exceed traditional methods, offering a more flexible approach to autonomous driving in dynamic and unpredictable environments.

A solar-powered electric motorcycle built in Africa completed a 6,000 km journey from Nairobi to Stellenbosch, demonstrating the real-world potential of clean mobility on the continent. The initiative, a collaboration between Stellenbosch University and Roam, tested the Roam Air motorcycle under African conditions, focusing on performance, battery efficiency, and solar charging over long distances. Beyond the journey itself, the project highlights a bigger goal: developing practical, locally suited electric mobility solutions and infrastructure for Africa. The trip concluded with the launch of Stellenbosch University’s Electric Mobility Lab, reinforcing the role of research, partnerships, and innovation in shaping the future of sustainable transport.

This research investigates how three autonomous vehicle control algorithms (Pure Pursuit, MPC, and MPCC) perform when moving from simulation to real-world testing on the F1Tenth platform. While MPCC performed best in simulation, real-world testing revealed that localisation errors, sensor noise, delays, and hardware limits significantly impacted performance. MPC was successfully adapted using filtering and delay compensation, while MPCC failed due to computational constraints. The key takeaway: bridging the sim-to-real gap requires more than good algorithms, it depends on reliable sensors, accurate localisation, and sufficient processing power.

This project examined whether reinforcement learning could reduce the energy used to cool telecom containers without allowing internal temperatures to drift into unsafe ranges. The result was a simulation-based control system that combined a fitted thermal model with a Deep Q-Network agent and delivered lower energy use than conventional automatic control strategies in both humid and arid conditions.

A final-year mechatronic engineering project at Stellenbosch University explored how speech technology and toy design can work together in early numeracy assessment. The result was an interactive toy car that asks simple math questions, listens to children’s spoken answers, and responds in real time using a low-resource digit recognition system built for child speech.