Semantic- and Density-Aware Planning for Accessibility-Preserving Multi-Object Placement

Benno Wingender; Nils Dengler; Nicolas Busch; Sicong Pan; Maren Bennewitz

cs.ROAug 17, 2026

Semantic- and Density-Aware Planning for Accessibility-Preserving Multi-Object Placement

Authors: Benno Wingender, Nils Dengler, Nicolas Busch, Sicong Pan, Maren Bennewitz

Abstract

Long-term manipulation planning requires robots to reason not only about immediate task success but also about how current decisions affect future interactions with the environment. In this context, household service robots may need to organize groceries in partially occupied shelves while using limited storage space efficiently and preserving access for subsequent placements. In this paper, we consider an online multi-object shelf-placement setting in which future objects arrivals are unknown. Existing approaches do not jointly address semantic organization, dense space utilization, and manipulator accessibility during sequential shelf filling. To address this gap, we propose Semantic-Dense Placement Planning (SDPP), an accessibility-preserving approach that ranks candidate poses using a semantic-density score combining inter-object semantic similarity with spatial proximity. An Accessibility Map (AM) further filters candidates unlikely to be reachable before motion planning and penalizes placements that reduce the remaining accessible workspace. Simulation experiments show that SDPP significantly improves semantic placement quality over state-of-the-art baselines and achieves the highest average shelf density, while the AM substantially reduces the time required to identify feasible placement poses. A qualitative real-world experiment demonstrates the applicability of our pipeline in a domestic shelf-storage scenario.

Figures & tables

Fig. 1 : Overview of the online multi-object shelf-placement task and the proposed approach. Starting from a partially occupied multi-level shelf, objects are observed and placed one at a time without knowledge of future arrivals. At each step, candidate placements on the selected shelf level are evaluated according to semantic-dense organization and accessibility preservation, after which the highest-ranked feasible placement is executed. This process is repeated until no further object can be placed, producing a dense and semantically organized shelf arrangement.

Fig. 2 : Overview of our semantic-dense placement planning pipeline. The perceived shelf environment is converted into 2D occupancy map for each shelf level (left), from which the Accessibility Map (AM) identifies occupied (gray) and manipulation-inaccessible (black) regions (green panel). Candidate placement poses are sampled in the remaining workspace and evaluated using a semantic-density score, which combines GloVe-based [ 24 ] object similarity with spatial proximity (orange panel), and a space-preservation penalty (blue panel) that estimates the loss of accessible shelf space caused by each placement pose. The combined score ranks candidate placements poses. pairs release and grasp pose candidates, and then sequentially checks for a collision-free pick-and-place trajectory (right).

Fig. 3 : Pairwise semantic similarity matrix computed from GloVe embeddings [ 24 ] after contrast normalization. Warmer colors indicate semantic similarity, while cooler colors denote dissimilarity.

Panel of Figure 4 — Fig. 4 : Evolution of shelf-arrangement metrics over successive placements, averaged over 10 paired trials. Each trial starts with 12 objects, denoted as placement step 0, and terminates when no executable placement can be found for the current object. Panels (a)–(c) report the semantic score, proximity score, and shelf density, respectively. SDPP maintains the strongest semantic organization as the shelf is filled while achieving shelf utilization comparable to the strongest density-oriented baselines.

Ablations
Variant	Add. Obj. Placed $\uparrow$	Semantic Sum $\uparrow$	Semantic $\uparrow$	Proximity $\uparrow$	Density $\uparrow$
Ours	$\mathbf{30.7\pm 3.9}$	$\mathbf{61.6\pm 11.5}$	$\mathbf{1.440\pm 0.180}$	$\mathbf{0.788\pm 0.036}$	$\mathbf{0.261\pm 0.016}$
SPS [ 13 ]	$4.7\pm 1.5$	$14.1\pm 1.9$	$0.855\pm 0.083$	$0.428\pm 0.095$	$0.100\pm 0.010$
SPS++ [ 13 ]	$28.2\pm 3.6$	$48.7\pm 10.0$	$1.208\pm 0.163$	$0.718\pm 0.041$	$0.245\pm 0.018$
Dengler [ 29 ]	$28.6\pm 4.6$	$32.4\pm 4.4$	$0.802\pm 0.065$	$0.557\pm 0.042$	$0.248\pm 0.018^{\dagger}$
Random	$19.9\pm 2.7$	$25.4\pm 3.1$	$0.799\pm 0.041$	$0.505\pm 0.047$	$0.193\pm 0.018$

TABLE I: Comparison of SDPP with the evaluated baselines and ablations. Each shelf initially contains 12 objects; Add. Obj. Placed denotes the number of additional objects successfully placed. All methods are evaluated over 10 paired trials using the same initial shelf configurations and incoming-object sequences. Values are reported as mean

\pm

standard deviation, and bold indicates the best result among the main methods. The semantic sum is computed over all objects in the final shelf, whereas the semantic, proximity, and density scores are averaged per object. Paired Wilcoxon signed-rank tests show significant differences between SDPP and each baseline for all reported metrics (

p<0.05

), except for shelf density compared with Dengler [ 29 ] , marked by † . No statistical significance tests are reported for the ablation variants. Among the main methods, SDPP achieves the strongest overall performance, combining the best semantic organization with the highest placement capacity and average shelf density. The ablations show that increasing the accessibility-preservation weight improves placement capacity and density, but eventually trades off semantic quality.

Fig. 5 : Mean accumulated IK and motion-planning time required to identify the first executable placement as the shelf is progressively filled, averaged over 20 runs. The no-filter baseline is shown in lime, while the AM-based strategies are shown in yellow (FC), pink (CC), and blue (FC $\rightarrow$ CC), where FC is followed by CC if no feasible pose is found. The red curve indicates the remaining accessible shelf area. AM-based filtering substantially reduces the candidate-search time, while FC $\rightarrow$ CC retains a less restrictive fallback when FC rejects all feasible candidates.

Explore similar work

Oct 8, 2026

Non-Prehensile Multi-Object Placement with Mobile Robots via Diffusion-Based Trajectory Planning

Non-prehensile manipulation with mobile robots has primarily targeted placement in coarse goal regions. Precision placement has been studied with manipulators on benchmarks such as Push-T, where an omnidirectional end effector can readily re-approach an object from any side. For a mobile robot, correcting the same placement errors may require wide repositioning maneuvers around the object. We formalize multi-object pushing with a mobile robot as a multi-target assignment problem at tight tolerances. We propose a diffusion planner with a causal factorization: each robot denoising step is followed by an object step conditioned on the current clean robot estimate. This factorization allows us to sample multiple object outcomes for a single robot plan, capturing the multi-modality of pushing. We also design a set of pushing-specific costs for mobile robots, used for both guidance and best-of-

N

sampling at inference. On Push-T, our planner attains the highest success rate of the methods we compare, while having the lowest rate of implausible object motion. Across three multi-box pushing environments resembling warehouse scenarios, it raises the fraction of boxes placed by

21.5

percentage points over an unguided baseline, and replans

1.8\times

faster than a planner that denoises the two trajectories together. Lastly, we deploy the guided planner on a TurtleBot 3 Burger with object shapes absent from the training data, where it places

85.0\%

of objects within tolerance against Diffusion Policy's

37.5\%

Lance Pharand, Megnath Ramesh, Mark Chiu +2

Electrical and Computer Engineering Department at the University of Waterloo, ON, Canada

Non-Prehensile Manipulation Mobile Robotics Diffusion-Based Planning

Ask PDF

Aug 7, 2026

Complete, Scalable, and Robust Prioritized Planning for Multi-Robot Ordered Storage and Retrieval at Maximum Capacity

Automated warehouses face a fundamental trade-off between maximizing storage density and achieving high retrieval throughput. While puzzle-based storage (PBS) architectures increase capacity by eliminating aisles, coordinating multiple robots in these high-density spaces is computationally challenging due to the potential for deadlocks. This paper introduces a novel multi-robot formulation for the ``ordered storage and retrieval problem at maximum capacity''. The focus is on rectangular grids accessible from a single boundary, where loads need to be first stored up to full capacity and then efficiently retrieved, given a planned departure sequence. This work bridges the gap between geometric feasibility and execution efficiency by leveraging the properties of relocation-free arrangements. These properties guide an online, prioritized multi-agent path-finding algorithm, which is the main contribution of this work. Unlike general centralized planners, the approach exploits the specific invariants of the storage arrangement to guarantee completeness and prevent deadlocks, enabling scalability. Experiments demonstrate that the method achieves near-linear improvement in makespan with respect to the number of robots, up to

m = C

, where

C

is the grid width. Crucially, the algorithmic overhead of supporting robustness is negligible; the system handles uncertainty in departure sequences using robust storage arrangements with no significant penalty in execution speed compared to the non-robust baseline.

William Zhang, Tzvika Geft, Jingjin Yu +1

Department of Computer Science, Rutgers University, New Brunswick, NJ, USA

Multi-Robot Motion Planning Multi-Robot Systems Robust MAPF Multi-Agent Path Finding

Ask PDF

Sep 17, 2026

Execution-Aware Pre-Execution Ranking for Grasp-Conditioned Robotic Placement

A geometrically valid placement can still be difficult to execute because the selected grasp changes the required end-effector pose, collision geometry, and transport motion. Placement is formulated as a pre-execution ranking problem in which supplied grasp-placement candidates are scored before planning. The model combines a typed target-conditioned point cloud with three pose descriptors and hierarchical heads for planning success and execution success conditioned on planning. On a 30-object, 1,235-scene dataset with scene-group-held-out splits, three-seed top-1 success on covered test groups reaches 85.63 +/- 1.08% for joint selection and 79.84 +/- 0.16% for fixed-target ranking. For the designated frozen seed-42 checkpoint, top-1 success improves from 72.84% to 85.78% over full-pool cuMotion for joint ranking and from 59.65% to 79.67% for fixed-target ranking. Frozen transfer to xArm7/MoveIt requires no xArm-specific retraining. Across 27 locked cases, 13 complete end to end (48.15%). Of the 16 cases that pass Top-5 preflight and begin execution, 13 succeed (81.25%). Candidate-level deployment-feasibility prediction reaches 81.25% recall, 85.20% specificity, and 83.23% balanced accuracy.

Tianyuan Liu, Rutherford Agbeshi Patamia, Benjamin Champion +2

Deakin University, Australia

Robotic Grasping Robotics Robotic Manipulation Robot Motion Planning

Ask PDF