Low-Resolution Perception for Robotic Packing

Giuseppe Fabio Preziosa; Federico Vignoni; Chiara Castellano; Marco Faroni; Andrea Maria Zanchettin; Paolo Rocco

cs.ROAug 26, 2026

Low-Resolution Perception for Robotic Packing

Authors: Giuseppe Fabio Preziosa, Federico Vignoni, Chiara Castellano, Marco Faroni, Andrea Maria Zanchettin, Paolo Rocco

Organizations: Politecnico di Milano, Piazza Leonardo da Vinci, 32. Milano (Italy).

Abstract

This work tackles the problem of scalable perception for robotic packing with low-cost, low-resolution depth sensing. We propose a framework where reconstruction cues drive next-view selection and grasp evidence updates a per-object stability estimate, jointly deciding what to acquire next and when to grasp. During the reconstruction, a low-resolution Next Best View (NBV) strategy explicitly avoids redundant views while preserving task-relevant geometry. We validate the approach in two steps: (i) an ablation study of the utility function under very low resolution, and (ii) a full end-to-end evaluation across policies, showing how low-resolution perception is a practical, scalable option for robotic packing.

Figures & tables

Figure 1: Pipeline overview. The Object Manager selects pick or view . For view , a target object is chosen (Exploration/Exploitation) and LR–NBV plans the next pose. The Grasp Generator outputs scored 6-DoF grasps. The loop repeats until all objects are removed.

Figure 2: Ray validity and Exploration gain. Low resolution leads to similar scores and redundant views. Image from Preziosa et al. (2025) .

Panel of Figure 3 — Figure 3: Acquisition results for four objects: horizontal box (top left), vertical box (bottom left), gripper (top right), and flange (bottom right). For each object: (a) estimated dimensions and IoU; (b) canonical dimensions; (c) surface coverage, number of poses, and useless poses. Image from Preziosa et al. (2025) .

sampled_poses	$w_{e}$	$w_{d}$	OctoMap_res[m]
50	500	0.05	0.01

Table 1: Study A parameters.

Object	R60–XPLR	R60–XPLT	R30–XPLR	R30–XPLT
Obj A	62.5%	70.8%	54.2%	54.2%
Obj B	66.7%	58.3%	83.3%	75.0%

Table 2: Grasp success

Object	R60–XPLR	R60–XPLT	R30–XPLR	R30–XPLT
Obj A	42.55%	56.29%	39.79%	44.36%
Obj B	54.77%	59.83%	50.45%	47.32%

Table 3: Mean reconstruction at grasp time

Figure 4: Failure breakdown by object and condition (percentages over fails)

Figure 5: (a,c) Number of acquisitions required before each grasp for three objects, under different acquisition and resolution modes. The x-axis denotes the grasp index (#1–#3). (b,d) Corresponding experimental setups for Obj A and Obj B.

Robotic Perception

Robotic perception enables robots to interpret scenes, objects, motion, and contact from cameras and other sensors to guide action.

See all 320 papers

UniScale: Unified Scale-Aware 3D Reconstruction for Multi-View Understanding via Prior Injection for Robotic Perception
Feb 26, 2026Mohammad Mahdavian, Gordon Tan, Binbin Xu +33D Reconstruction Multi-View 3D Reconstruction
Huawei Noah’s Ark Lab · University of Toronto
Pickalo: Leveraging 6D Pose Estimation for Low-Cost Industrial Bin Picking
Apr 6, 2026Alessandro Tarsi, Matteo Mastrogiuseppe, Saverio Taliani +2Robotic Grasping Object Pose Estimation
MESH Facility (formerly iCub Tech), Istituto Italiano di Tecnologia, Genova 16163, Italy · Institut des Systèmes Intelligents et de Robotique (ISIR), Paris 75005, France · Generative Bionics, Genova 16152, Italy +1
Behavior Cloning for Active Perception with Low-Resolution Egocentric Vision
May 13, 2026Anthony Bilic, Chen Chen, Ladislau BölöniActive Vision for Robot Manipulation Active Sensing
Department of Computer Science, University of Central Florida, Orlando, FL, USA

Feb 26, 2026J/K move · Enter open · S save

UniScale: Unified Scale-Aware 3D Reconstruction for Multi-View Understanding via Prior Injection for Robotic Perception

Mohammad Mahdavian, Gordon Tan, Binbin Xu, Yuan Ren+2

Huawei Noah’s Ark Lab · University of Toronto

3D Reconstruction Multi-View 3D Reconstruction Feed-Forward 3D Reconstruction Robotic Perception Camera Pose Estimation

Open paper AskPDF arXiv

We present UniScale, a unified, scale-aware multi-view 3D reconstruction framework for robotic applications that flexibly integrates geometric priors through a modular, semantically informed design. In vision-based robotic navigation, the accurate extraction of environmental structure from raw image sequences is critical for downstream tasks. UniScale addresses this challenge with a single feed-forward network that jointly estimates camera intrinsics and extrinsics, scale-invariant depth and point maps, and the metric scale of a scene from multi-view images, while optionally incorporating auxiliary geometric priors when available. By combining global contextual reasoning with camera-aware feature representations, UniScale is able to recover the metric-scale of the scene. In robotic settings where camera intrinsics are known, they can be easily incorporated to improve performance, with additional gains obtained when camera poses are also available. This co-design enables robust, metric-aware 3D reconstruction within a single unified model. Importantly, UniScale does not require training from scratch, and leverages world priors exhibited in pre-existing models without geometric encoding strategies, making it particularly suitable for resource-constrained robotic teams. We evaluate UniScale on multiple benchmarks, demonstrating strong generalization and consistent performance across diverse environments. We will release our implementation upon acceptance.

Abstract

Robotic Perception

Explore similar work

UniScale: Unified Scale-Aware 3D Reconstruction for Multi-View Understanding via Prior Injection for Robotic Perception

Pickalo: Leveraging 6D Pose Estimation for Low-Cost Industrial Bin Picking

Behavior Cloning for Active Perception with Low-Resolution Egocentric Vision