NexusPlace

NexusPlace is a Hybrid AI capability providing the perception and decision layer required to detect, locate and place physically constrained objects within bounded spaces. It combines computer vision, spatial reasoning and optimisation-driven planning to transform raw scene data into safe, efficient and explainable placement actions for robotic and industrial systems.

Unlike pure learning-based approaches that treat placement as a black-box prediction, NexusPlace pairs deep-learning perception with symbolic constraint solvers — ensuring that every placement decision respects physical geometry, safety margins, sequencing dependencies, and operational rules. The result is an AI system that an operator can trust, inspect, and override, backed by NexusTrust assurance and orchestrated through NexusFlow when part of a wider industrial workflow.

---

Key Features

• Computer-vision-based scene understanding — object detection, pose estimation, and spatial mapping from multi-camera and depth sensor inputs
• Spatial reasoning over bounded environments — understanding workspace geometry, occupied volumes, and available placement zones
• Optimisation-driven placement planning — constraint solvers that find safe, efficient configurations for packing and assembly tasks
• Constraint-aware decision-making — respecting safety margins, geometric tolerances, sequencing dependencies, and operational rules
• Explainable placement rationales — every action plan includes the reasoning chain so operators understand why objects are placed where they are
• Robotics integration — standard messaging and control interfaces for industrial robotic arms and automated systems

---

Application Areas

---

Technical Specification

AI Approach	Hybrid: deep learning perception + symbolic / optimisation reasoning
Perception	Multi-camera, depth sensor, and LIDAR-compatible inputs
Planning	Constraint solvers for placement, packing, and sequencing problems
Compute	GPU inference for vision; CPU for optimisation and constraint solving
Output	Action plans with explainability metadata and confidence scores
Integration	NexusFlow orchestration, NexusTrust assurance, standard robotics interfaces
Cluster	Cluster 4 — AI, Data & Robotics

---

How It Works

Scene Perception

NexusPlace begins by building a real-time understanding of the workspace. Multi-camera and depth sensor inputs are fused into a 3D scene model that identifies every object in the environment — its type, position, orientation, and bounding geometry. Object detection models trained on domain-specific datasets recognise components, containers, tools, and obstacles. Pose estimation determines how each object is oriented in space, which is critical for assembly tasks where insertion angle and rotational alignment matter.

Spatial Reasoning

With the scene understood, the spatial reasoning layer maps out the available placement zones within the bounded workspace — accounting for occupied volumes, clearance requirements, reach constraints of the robotic arm, and any no-go zones defined by safety rules. This is not a simple collision check; it maintains a volumetric model of the workspace that updates in real-time as objects are added, moved, or removed, allowing NexusPlace to reason about what configurations are physically possible and what the consequences of each placement would be.

Constraint-Aware Planning

The planning engine takes the perceived scene and the spatial model and solves for the optimal placement configuration. This is where the hybrid AI architecture pays off: deep learning handles the perceptual ambiguity of real-world scenes (reflections, occlusions, novel objects), while symbolic constraint solvers handle the combinatorial complexity of packing and sequencing. Constraints include geometric fit, weight distribution, assembly sequence dependencies, safety margins, and operational preferences. The solver produces a ranked set of placement plans, each annotated with a feasibility score, risk assessment, and explainability rationale.

Explainable Action Plans

Every placement plan NexusPlace generates includes a human-readable rationale: why this position was chosen over alternatives, which constraints were binding, what the confidence level is, and what would need to change for a different outcome. This explainability is not decorative — in manufacturing environments where a misplaced component can cause downstream failures, operators need to understand and trust the AI’s reasoning before authorising execution. When paired with NexusTrust, these rationales feed into a continuous assurance record that tracks placement accuracy, failure modes, and system reliability over time.

Robotics Execution

Once a placement plan is approved — either automatically for routine operations or by an operator for novel or high-risk scenarios — NexusPlace translates the plan into robotic control commands through standard messaging interfaces. It supports integration with major industrial robotic platforms, providing waypoint trajectories, gripper commands, and real-time feedback loops that adjust execution if the object or environment shifts during handling. The system monitors execution against the plan and flags deviations for review.

---

The Hybrid AI Advantage

Most robotic placement systems choose either learning-based approaches (flexible but opaque, brittle on edge cases) or classical planning (precise but rigid, slow to adapt). NexusPlace combines both:

• Deep learning handles perception — recognising objects in cluttered, variable-lighting, real-world conditions where hand-coded vision fails
• Symbolic reasoning handles planning — guaranteeing that physical constraints, safety rules, and sequencing logic are respected, not approximated
• Optimisation handles efficiency — finding configurations that minimise cycle time, maximise space utilisation, or balance multiple competing objectives
• Explainability bridges the gap — operators see both what the AI perceives (vision) and why it decided what it decided (reasoning), building the trust needed for human-robot collaboration

This hybrid approach is particularly valuable in Industry 5.0 contexts where human-centric manufacturing requires AI that works alongside people — not as a black box, but as a transparent, inspectable decision-making partner.

---

Heritage & Lineage

NexusPlace leverages VTG’s contribution to the ULTIMATE project — specifically the toolkit for Robotic AI-Based Data Analysis and Visualisation, published at ACIIDS 2024 (Springer LNCS). The research demonstrated how hybrid AI architectures combining deep-learning perception with constraint-based planning could achieve reliable, explainable object placement in bounded industrial environments — the core capability now productised as NexusPlace.

Target Calls: Horizon Europe Cluster 4 (AI, Data and Robotics), Made in Europe partnership, manufacturing automation calls, and Industry 5.0 topics requiring human-centric robotic AI.