Joshua Knights

Commonwealth Scientific and Industrial Research Organisation

Papers

Total Citations

H-Index

About

Joshua Knights is a researcher advancing the frontiers of robotic perception and self-supervised learning, with a primary focus on place recognition and video representation. His most impactful work, "InCloud: Incremental Learning for Point Cloud Place Recognition" (26 citations), tackles a critical challenge in robotics: the severe performance degradation of deep learning models when deployed in unseen or dynamic environments. By introducing an incremental learning framework, Knights enables robots to continuously adapt without catastrophic forgetting, a key contribution for long-term autonomous operation. He further addresses reliability in "Uncertainty-Aware Lidar Place Recognition in Novel Environments" (5 citations), pioneering methods to quantify model confidence in unfamiliar settings—a vital step toward trustworthy deployment. In the domain of video understanding, his paper "Temporally Coherent Embeddings for Self-Supervised Video Representation Learning" (6 citations) introduces a novel approach that explicitly enforces temporal consistency in embedding spaces, moving beyond traditional ranking-based methods. Collectively, Knights’ work bridges the gap between robust place recognition and efficient representation learning, with his incremental and uncertainty-aware techniques poised to influence next-generation autonomous systems operating in complex, real-world environments.