Innovation Spring 2026

“

CNNs learn hierarchical representations across different “layers” of processing. Early layers can capture factors like local texture and shape cues, while deeper layers encode higher-level structure such as object boundaries and relative size. This approach resolves many of the issues encountered in classical image processing. For example, the EP signals captured by Kim’s prototype sensing robot consists of small voltage fluctuations distributed across multiple sensing sites on a leaf. These fluctuations do not act independently. “Plants are living, fluid-based organisms, but they don’t have any strong mechanical or electrical signal generating organs,” noted Kim. CNNs compensate for that trait by finding patterns between the raw inputs to create coherent estimates that take interdependency into account during processing. Time data introduces an additional element that makes these systems more accurate and useful. Vision systems combine space and time, allowing algorithms to predict tasks such as estimating time to harvest. This data can also be useful throughout the whole

We’ve been talking to our customers, and growers are excited about the potential that this type of technology could have. For the industry, in my view, this type of development could be a game changer.

Blake Ponuick Co-founder and CEO, Nourish Labs

”

growing process. At 4AG, intelligent automation extends throughout the growth cycle. “To pick the whole bed, you actually have to manage it from the start,” Gibson explained. “We identify small pins as they emerge, classify primary mushrooms, and remove anything that will interfere with growth later.” A new approach to growing These systems, in addition to optimizing setups, are also fundamentally changing how growers think about the farming process. For example, the trade-off between harvesting time constraints and maximizing crop yield has traditionally been a significant issue for mushroom cultivators, since mushrooms grow rapidly. “Humans do two eight-hour shifts and leave overnight,” said Gibson. “Mushrooms double in size every 24 hours. A human might see a 40-mm mushroom at the end of a shift and pick it so it doesn’t oversize by morning. Our robots see the mushroom, know they’ll be back in an hour, and wait.” That ability to return repeatedly, without rest or interruption, changes the economics of harvesting. Over time, “We optimize yield and get more energy out of the same amount of compost,” Gibson said. For the industry, introducing robotic harvesting could relieve growers from labour shortages. Harvesting requires long hours of precise labour, making it strenuous and high-turnover; according to the

20

Spring 2026

Innovation