We investigate artificial intelligence, computer vision, sensing, geospatial data and robotics to develop systems capable of observing, understanding and acting on real-world operations. Convex research connects models, data and field experimentation to the engineering behind Nexa M1, Nexa Vision, Atmos, Synapse Control and Axis.
Detection, segmentation, classification and scene understanding to transform imagery from satellites, drones, robots and fixed cameras into structured operational information.
Research on time series, multispectral imagery, spectral indices, spatial change and territorial data integration for continuous understanding of large areas.
Combining imagery, weather data, sensing, historical context and language to produce more robust interpretations of the physical environment.
Perception, planning, control and task execution for robotic systems operating in agricultural, industrial and other physical environments.
Models for interpreting vegetation vigor, stress, spatial variability, environmental conditions and other signals relevant to agricultural monitoring.
Research into architectures that bring perception and decision-making closer to sensors and robots, reducing reliance on remote processing when response time is critical.
Every research area runs through the same engineering loop, from raw field capture to operation that feeds new data back into the next iteration.
Imagery, sensors, environmental data and operational context.
Organization, annotation, cleaning and preparation of data.
Model development and refinement.
Quantitative metrics and error analysis.
Verification in scenarios representative of the real problem.
Models incorporated into Convex products.
Real usage generating new data for further iterations.
Each problem calls for its own family of metrics. The lists below describe the indicators we evaluate against — not published results.
Research directions never sit in isolation: each one feeds the engineering of a product in the ecosystem.
Ongoing work, described as research directions rather than completed technology.
Combining imagery, sensing and context to reduce ambiguity when interpreting real scenes.
Investigating the consistency of spectral indices over time across continuously monitored areas.
A research direction on spatial variability and stress signals in crops.
Perception, planning and task execution in unstructured physical environments.
Turning technical diagnostics into traceable recommendations for people working in the field.
Architectures that connect capture, decision and execution within a single loop.
Building models for the physical world depends less on raw volume than on diversity: different crops, terrain, sensors, resolutions, seasons, and lighting and cloud conditions.
We therefore treat data as a research object in its own right. Annotation, quality control, train/validation separation, error analysis and reproducibility are as much engineering work as model development.
Technical publications will be made available here as research work is completed.
Convex is open to applied research projects, technology validation, field studies and collaboration with universities, research centers and organizations working in artificial intelligence, agriculture, sensing and robotics.
Discuss research →Data before opinion.
Metrics before promises.
Validation before scale.
Engineering for the real world.
