Michael Mistry among five new fellows of the Bayes Innovation Fellows programme

Michael Mistry was competitively selected from across the university and will benefit from individual coaching, resources, time and seed funding to develop his commercial endeavours. 

Michael is focused on tackling one of the most pressing challenges in the field of Robotics: enabling robots to perform complex and highly interactive tasks with precision and reliability. While industries such as manufacturing and fulfilment have made strides in integrating robotic manipulation, tasks that involve intricate handling—like material handling, picking, packing, and assembly—remain difficult. The problem stems from the advanced sensing and control required to execute these tasks without causing damage to the robots, their surroundings, or, most importantly, humans.

His vision is to enhance robotic manipulation by making it faster, more affordable, and more reliable through the use of data-driven predictive modeling and control. The recent rise of deep learning has shown that crucial features can be automatically extracted from data, particularly in visual domains like image classification. However, for contact-rich tasks, it is often the interaction between multiple sensory inputs—such as vision, touch, and kinematics—that reveals the most valuable information.

To address this, Michael will develop models that learn a compact, action-orientated latent representation of multi-modal input. The models will be trained to reproduce the rich sensory consequences of robot action, even when using a reduced (lower-cost) sensor suite. These models can then be used to detect anomalies in real-time, pre-empt failures and automatically annotate outcomes. The models will also be flexible,  adjusting as things change and sharing what they learn across different systems.

Michael acknowledges that the initial phase of data collection and training of such models might involve installing a range of additional sensors—such as cameras, tactile sensors, force/torque sensors, and microphones—some of which may be redundant. However, once trained, the models will identify the minimum necessary sensors needed, leading to significant cost savings when used in the field. Moreover,  the models can act as a digital twin (if specifically trained for this), allowing for virtual testing and cost-effective trials with different setups or combinations.

The year-long Bayes Innovation Fellows programme launched last year is designed to develop a translational mindset amongst early-stage academics and promote entrepreneurism and innovation activities across the university. 

Francesco Tudisco (School of Mathematics), Gary Watmough (School of Geosciences), Giovanni Stracquadanio (School of Biological Sciences) and Shiwei Wang (School of Engineering) have also been selected as Bayes Innovation Fellows 2024.

More about Michael’s research and innovation idea

Bayes Innovation Fellows Cohort 2024

Bayes Innovation Fellows Website