Object Intelligence Model: Enabling Context-Aware Robotic Interaction Through Integrated Object Detection and Material Recognition

Robots operating in human-centric environments, such as homes, hospitals, or disaster zones, demand perception capabilities that extend beyond simple object identification to ensure intelligent and safe interactions. Current object detection systems excel at labeling objects but fail to infer critical physical and functional properties, such as material, weight, or contextual utility, limiting their effectiveness in dynamic settings. This research introduces the Object Intelligence Model (OIM), a deep learning framework designed to endow robots with human-like reasoning about objects. For instance, when a mobile phone is about to fall, a human instinctively recognizes it, understands its glass screen's fragility, and acts swiftly to catch it. OIM mirrors this by detecting the phone, inferring its material properties, and prioritizing immediate catch action. OIM integrates real-time object detection, a static property database, dynamic material analysis using convolutional neural networks (DenseNet121 and DenseNet169) trained and tested on a subset of 3500 images from the MINC-2500 dataset, and context-aware decision-making to assess properties like fragility or task suitability. Preliminary results from material recognition models show promising accuracy despite limited resources, with comparisons to baseline models highlighting OIM's potential achieving an 81.41 % train accuracy and 72.00 % test accuracy with DenseNet169. Applicable to assistive robotics, disaster response, and logistics, OIM enhances robotic autonomy and safety by bridging passive recognition and active reasoning. From recognition to reasoning: Robots that think like humans.