Answers are grounded in this paper and cite their sources.
Reliable harvesting requires visual information that remains useful for planning and execution. This review examines four complementary functions of harvesting robot perception: detection and segmentation (D), metric three-dimensional localization (L), pose and task-relevant geometry (P), and environment reconstruction (E). D identifies the target, L locates it in a metric coordinate frame, P supplies the geometry needed for approach and detachment, and E describes the surrounding scene and its spatial context. A complete harvesting process requires these functions to work together, although individual studies may cover only some of them. The review distinguishes complete-system evidence, harvesting-specific module evidence, and adjacent evidence, using reported configurations to show how these functions are combined. An actionability gate assesses target identity, coordinate consistency, task-relevant geometry, scene coverage, uncertainty, information age, and the downstream planning or control consumer. A three-level framework connects module performance with information retained across interfaces and with robot or field deployment. By making the perception-to-action pathway explicit, this review clarifies the visual information required at each decision interface, identifies evidence gaps, and provides an engineering basis for designing and evaluating deployable harvesting-robot systems.
No takes yet. Share an insight, caveat, or question.
Gao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: