RoboCup Smart Manufacturing League

• RoboCup Smart Manufacturing League is an industrial benchmark integrating mobile manipulation, autonomous control, and task execution in standardized factory settings.
• The league combines a traditional RoboCup@Work-style benchmark for pick-and-place tasks with a broader modular vision encompassing logistics, assembly, and human–robot collaboration.
• It emphasizes holistic system integration with precise metrics for mapping, navigation, perception, and manipulation to drive innovation in smart manufacturing.

Searching arXiv for papers on RoboCup Smart Manufacturing League and related RoboCup industrial leagues. arxiv_search(query="RoboCup Smart Manufacturing League RoboCup@Work RoboCup Logistics League industrial league", max_results=10, sort_by="relevance") Searching for the specific arXiv IDs mentioned and closely related league papers. arxiv_search(query="(Khalili et al., 2024) OR (Dissanayaka et al., 15 Jul 2025)", max_results=10, sort_by="relevance") The RoboCup Smart Manufacturing League denotes a RoboCup competition framework for industrial robotics centered on autonomous operation in manufacturing environments. Recent literature uses the term in two closely related senses: as an international benchmark for industrial-style mobile manipulators, explicitly described as “formerly known as RoboCup@Work,” and as a broader, newer industrial league vision that extends beyond production logistics to encompass flexible assembly, human–robot collaboration, and humanoid robotics (Khalili et al., 2024, Dissanayaka et al., 15 Jul 2025). In both senses, the league functions as a benchmark for integrated perception, planning, control, and task execution under standardized but semi-structured factory conditions.

1. Historical positioning and nomenclature

In the 2024 technical report on Auriga’s mobile manipulator, the RoboCup Smart Manufacturing League is identified as the continuation of RoboCup@Work and is characterized as an international benchmark for industrial-style mobile manipulators operating in semi-structured “shop-floor” environments (Khalili et al., 2024). The emphasis in that account is on end-to-end robot competence within a compact arena, with tasks such as pick-and-place, kitting, and sorting/inspection.

A 2025 vision paper uses the same league name in a broader institutional sense. It starts from the RoboCup Logistics League, which had focused on task planning, job scheduling, and multi-agent coordination in a smart factory scenario, and argues that the competition should be re-envisioned as a larger smart manufacturing scenario composed of several tracks that are initially independent but gradually combined into one smart manufacturing scenario (Dissanayaka et al., 15 Jul 2025). The proposed tracks cover production logistics, flexible assembly, human–robot collaboration, and humanoid robotics.

This dual usage is the main source of terminological ambiguity. A common misconception is to treat the name as referring to a single fixed ruleset. The literature instead indicates an ongoing transition: one usage denotes the established RoboCup@Work-style benchmark, while the other denotes an expanded league architecture intended to absorb production logistics and additional industrial-robotics challenges. This suggests that “Smart Manufacturing League” is best understood as both a current benchmark label and an evolving organizational umbrella.

2. Objectives and benchmark philosophy

The RoboCup@Work-derived formulation emphasizes two objectives: encouraging end-to-end solutions that integrate mapping, navigation, manipulation, and perception in industrial service tasks, and providing a reproducible, standardized arena so that different teams’ solutions can be fairly compared (Khalili et al., 2024). The benchmark is therefore not confined to a single subsystem. It rewards system integration across locomotion, localization, object recognition, grasp execution, and fault-tolerant behavior in repetitive or hazardous environments.

The broader 2025 vision articulates a twofold objective for the new league structure: increasing accessibility and lowering the entry barrier by dividing the competition into modular, self-contained tracks, and reflecting a wider spectrum of real-world industry-relevant tasks by eventually recombining these tracks into a single, large-scale smart-manufacturing scenario (Dissanayaka et al., 15 Jul 2025). This modularization is intended to let newcomers and specialists enter at the level of logistics and scheduling, flexible assembly, human–robot collaboration, or humanoid manipulation, while preserving a path toward full integration.

Taken together, these formulations locate the league at the intersection of benchmark design and research agenda setting. In the narrower benchmark, integrated autonomy for a mobile manipulator is the core problem. In the broader vision, integrated autonomy remains central, but the benchmark boundary expands to include online order arrivals, resource contention, collaborative safety constraints, and whole-body control.

3. Arena configurations and task repertoire

The literature describes both a compact standardized arena for the RoboCup@Work-style benchmark and a multi-track set of environments for the broader Smart Manufacturing League vision (Khalili et al., 2024, Dissanayaka et al., 15 Jul 2025).

Configuration	Environment	Core tasks
RoboCup@Work-style benchmark	3 m × 4 m modular panels; surrounded by 50 cm tall walls; up to four service stations with standardized fixtures; AprilTags or ARTags denote workspace origins and goal areas	Pick-and-place; kitting; sorting/inspection
Production Logistics track	4 m × 3 m modular floor with seven reconfigurable station modules; AGVs navigate marked lanes with QR-based floor localization	Fetch raw materials; deliver subassemblies; handle machine breakdowns or oversubscription
Flexible Assembly track	Bench-top arena; 3 × 2 grid of work zones; modular assembly fixtures and parts bins; vision markers for camera-based part localization	Pick and assemble standardized component kits; verify structural integrity; deliver completed assemblies
Human–Robot Collaboration track	2 m × 2 m workspace; human-only, robot-only, and shared zones; light-curtain scanners; tablet interface for higher-level commands	Co-operative kitting; shared-workspace assembly; collaborative quality inspection
Humanoid Robotics track	1.5 m × 1 m elevated platform; stairs, small conveyors, and scaled assembly stations	Dexterous manipulation; dynamic obstacle negotiation; high-level decision making

Within the RoboCup@Work-style setting, typical challenge scenarios include a static environment with known obstacle layout for initial mapping, a semi-dynamic setting with an “operator” occasionally moving a workpiece, and multi-task rounds in which the robot must sequence pick, transport, and inspection steps within 10 minutes (Khalili et al., 2024). In the broader league vision, the environment is partitioned into specialized tracks whose eventual recombination is itself a design goal (Dissanayaka et al., 15 Jul 2025).

The contrast is methodologically important. The compact arena stresses tightly coupled mobile manipulation under controlled reproducibility. The track-based proposal expands the benchmark surface to cover smart-factory functions that were previously separated across RoboCup subcommunities.

4. Rules, scoring, and formal evaluation models

In the RoboCup@Work-style benchmark, each correctly executed operation scores 10–20 points depending on difficulty, remaining time in a 600 s run converts at 0.1 points/s, and penalties apply for collisions (–5 points each), misplacement (–10 points), and exceeding workspace (–2 points) (Khalili et al., 2024). The scoring scheme directly couples correctness, efficiency, and safety-relevant failure modes. Because the arena is standardized, these point allocations turn manipulation accuracy, navigation reliability, and execution speed into comparable performance signals across teams.

The broader Smart Manufacturing League proposal retains a common scoring structure across tracks. It defines delivered tasks $N$, weighted complexity $C = \sum_{i=1}^{N} \omega_i$, and tardiness penalty $L = \sum_{i=1}^{N} \max(0, t_i - d_i)$, with overall score

$S \;=\; \alpha\,N \;+\; \beta\,C \;-\; \gamma\,L$

and empirically chosen weights $\alpha=5$, $\beta=2$, and $\gamma=1$ (Dissanayaka et al., 15 Jul 2025). In the logistics track, an additional utilization metric $U$ penalizes idle AGV time and contributes to a bonus term $\delta\,U$. Quality indices, including the fraction of assemblies passing an automated inspection or the rate of near-misses in human–robot tasks, are also tracked, but do not directly affect $S$ in early seasons.

The formalization is more explicit in the 2025 proposal than in the RoboCup@Work-style rules. For the logistics track, jobs $C = \sum_{i=1}^{N} \omega_i$0 are assigned to machines $C = \sum_{i=1}^{N} \omega_i$1 and AGVs $C = \sum_{i=1}^{N} \omega_i$2 via a time-indexed mixed-integer program, while the assembly track is modeled as a POMDP with state $C = \sum_{i=1}^{N} \omega_i$3, actions $C = \sum_{i=1}^{N} \omega_i$4, and policy

$C = \sum_{i=1}^{N} \omega_i$5

These formulations indicate that the league is not only an engineering competition but also a venue for explicit optimization and decision-theoretic models (Dissanayaka et al., 15 Jul 2025).

5. Technical methodologies exercised by the league

A concrete instance of the RoboCup@Work-style benchmark is provided by Auriga’s entry from Shahid Beheshti University, whose robot combines a 4-wheel Mecanum base, a telescopic 5-degree-of-freedom manipulator arm, a custom 3D-printed gripper, custom ESP32-based control boards, and an Nvidia Jetson Nano running ROS, SLAM, planning, and vision (Khalili et al., 2024). The base uses standard inverse and forward kinematics for omnidirectional motion, the manipulator’s forward kinematics is expressed as $C = \sum_{i=1}^{N} \omega_i$6, and reachability analysis via Monte-Carlo over joint limits yields a workspace radius of approximately 30 cm from the base axis and a vertical range of 20–60 cm.

Auriga’s onboard stack is representative of the integrated autonomy problems that the league is intended to stress. Mapping uses Hector SLAM on a 2D occupancy grid with scan matching and iterative Gauss–Newton or Levenberg–Marquardt updates, publishing a 0.05 m resolution grid on ROS’s “/map” topic (Khalili et al., 2024). Global navigation uses A* over a weighted occupancy-grid graph with total cost $C = \sum_{i=1}^{N} \omega_i$7, where the safety term depends on distance to the nearest occupied cell. Vision uses Ultralytics YOLOv5 (small variant), trained on a custom dataset of 2 000 images over 5 part classes, with final $C = \sum_{i=1}^{N} \omega_i$8 and $C = \sum_{i=1}^{N} \omega_i$9. Detected 2D boxes are fused with D435i depth to compute 3D centroids, and PCA on the segmented depth patch estimates object orientation about the vertical axis. Collision avoidance uses a complementary EKF combining wheel odometry with IMU yaw/acceleration and LiDAR scan-matching updates, while a dynamic occupancy grid at 10 Hz triggers an emergency stop if any occupied cell enters a 0.3 m safety radius.

The broader league proposal enlarges the technical scope beyond single-robot mobile manipulation. It highlights multi-agent coordination under online order arrivals and resource contention, dynamic job scheduling, fault diagnosis and re-planning with minimal human input, robust perception and manipulation in cluttered, unstructured environments, safe and intuitive human–robot interfaces, and whole-body control and balance for bipedal humanoids in tight-space navigation (Dissanayaka et al., 15 Jul 2025). A plausible implication is that the benchmark family is migrating from integrated autonomy at the robot level toward integrated autonomy at the factory-system level.

6. Empirical performance, limitations, and future trajectories

Auriga’s experimental evaluation in the standard @Work arena consisted of ten trials and reported an average first-mapping time of $L = \sum_{i=1}^{N} \max(0, t_i - d_i)$0, localization RMSE versus ground-truth AprilTags of $L = \sum_{i=1}^{N} \max(0, t_i - d_i)$1 with $L = \sum_{i=1}^{N} \max(0, t_i - d_i)$2, path planning time of $L = \sum_{i=1}^{N} \max(0, t_i - d_i)$3 at 5 cm grid resolution, pick-and-place cycle time of $L = \sum_{i=1}^{N} \max(0, t_i - d_i)$4 per object, and an overall task success rate of 90% with 9/10 trials delivering all 4 parts correctly (Khalili et al., 2024). The report also identifies practical lessons: tightly synchronized sensor fusion at 100 Hz is crucial for safe omnidirectional motion near obstacles, on-the-fly replanning improved robustness in semi-dynamic scenarios, and the 3D-printed gripper required post-print annealing to avoid finger fracture under repeated loading.

The same report states three principal limitations: a limited arm payload of 0.5 kg, the inability of current 2D SLAM to detect overhung obstacles such as suspended pipes, and occasional YOLO misclassification of small parts in low-light conditions (Khalili et al., 2024). Its proposed future improvements are to integrate 3D collision checking via the RealSense point cloud, upgrade to a 6-DOF wrist for in-hand manipulation and dexterous assembly, explore transformer-based detectors under varying lighting, and implement time-bounded anytime path planners such as RRT* and CHOMP.

At the league-design level, the 2025 proposal specifies infrastructure intended to ensure reproducibility and a level playing field: standardized robot platforms such as an AGV platform, a 6–7 DoF manipulator on a linear rail, a collaborative robot for the HRC track, and a low-cost humanoid; ROS 2 Foxy for inter-agent messages, OPC-UA for station control, MQTT for high-level order announcements; Gazebo with factory-module plugins, containerized Docker images for continuous integration, and a cloud-based leaderboard; and physical testbeds including a modular table system, safety-rated scanners, and a stair/platform kit (Dissanayaka et al., 15 Jul 2025). This suggests a future in which benchmark reproducibility, simulation-to-competition continuity, and gradual integration across tracks are treated as first-order design constraints.

The league’s significance, in both its current and proposed forms, lies in how it operationalizes smart-manufacturing research. In the RoboCup@Work-style benchmark, the unit of evaluation is the integrated mobile manipulator executing industrial service tasks under standardized conditions. In the broader Smart Manufacturing League vision, the unit of evaluation expands to a modular but ultimately unified factory scenario spanning logistics, assembly, collaboration, and humanoid operation.