Digital Model-Based Optimization of Mobile Manipulators for Human-Robot Collaborative Tasks

Digital Model-Based Optimization of Mobile Manipulators for Human-Robot Collaborative Tasks Christian Cella, Marco Faroni, Andrea M. Zanchettin, Paolo Rocco Abstract— We present an optimization framework for mobile manipulators in human-robot collaborative tasks integrating layout planning, base pose selection, task sequencing, and human-aware adaptation. A Particle Swarm Optimization scheme balances efficiency and safety by considering cycle time, manipulability, and operator proximity. A case study in box packing demonstrates the framework’s ability to minimize cycle times while ensuring seamless adaptation to human presence. I. INTRODUCTION Human-robot collaboration (HRC) is transforming industrial automation, enabling safer and more efficient workplaces where humans and robots share tasks dynamically. Mobile manipulators are increasingly deployed for operations requiring both reach and flexibility, yet their efficiency critically depends on optimal robot decision making (e.g., path planning and scheduling) and smart layout design, especially when human interactions are involved [1]. To address these challenges, we propose a framework for optimizing robot base positions and task scheduling during the predeployment phase. Although simulation-based methods are common in robotics, their application to optimizing mobile HRC processes during the pre-deployment stage remains an underexplored area. Our framework aims at filling this gap by jointly optimizing robot base positions and task sequences while balancing multiple conflicting objectives via iterative simulations on a black-box process model. We solve the resulting problem by combining Particle Swarm Optimization (PSO) [2] and a digital model to simulate the process and evaluate a set of key performance indicators (KPIs). We demonstrate its effectiveness in a box-packing scenario, with a robot on a linear guide and a human following a predefined schedule, that we assume being imposed by an external orchestrator. Videos of the simulations are available in the accompanying video.1Please not that the results described in this abstract are adapted from [3]. II. APPROACH We consider the optimization of mobile manipulators in human-robot collaborative environments, where robots perform pick-and-place operations while sharing the workspace with human operators. Such a problem is examplified in This study was partially carried out within the MICS (Made in Italy – Circular and Sustainable) Extended Partnership and received funding from Next-Generation EU (Italian PNRR – M4 C2, Invest 1.3 – D.D. 1551.1110-2022, PE00000004). CUP MICS D43C22003120001. The authors are with Politecnico di Milano, Piazza Leonardo da Vinci, 32. Milano (Italy) [email protected] 1The video is available at: https://youtu.be/Q5XaGDUOH0M Fig. 1: An example of base positioning and scheduling for a mobile manipulator. Fig. 1: The goal is to determine both the sequence of robot base positions and the order of tasks in a way that maximizes efficiency while ensuring operator safety. Human tasks are assumed to be known in advance from an Enterprise Resource Planning (ERP) system, while the robot tasks are generated iteratively through the proposed method. Since the optimization problem cannot be expressed in closed form, it is treated as a black-box and solved by employing a digital model of the collaborative environment. The proposed framework consists of the following main modules. A. Packing layout The first stage of the method determines where objects should be placed inside the available boxes. To do so, we use a modified version of the best-fit bin-packing algorithm [4]. This provides both the feasible placement positions and the capacity of each box, which together define the set of “place-side” targets for the robot. B. Base pose optimization Once the packing layout is defined, the next step is to determine suitable robot base positions for each pick-andplace task. We employ Particle Swarm Optimization (PSO) to search for candidate base poses. Each candidate is evaluated according to multiple performance indicators, such as execution time, robot manipulability, and feasibility. Infeasible solutions, for instance those leading to collisions, are penalized. The swarm then updates candidate poses through an iterative process that balances exploration and exploitation, eventually converging toward optimal base poses. C. Task sequencing and travel time With feasible base poses identified, the framework determines the order in which tasks should be executed. The strategy is based on minimizing the travel cost of the mobile base between consecutive tasks. At each step, the next operation is chosen as the one requiring the shortest 2025 I-RIM Conference October 17-19, Rome, Italy ISBN: 9788894580570 10.5281/zenodo.17629794 163 Fig. 2: a) Example of a Gantt chart generated by the proposed optimizer. b) Layout considered in the use case. additional travel. This rule is applied recursively, resulting in an efficient sequence of robot operations that reduces unnecessary movement and cycle time. D. Human-aware adaptation The final stage ensures safe human-robot interaction. The robot continuously adjusts its speed and acceleration depending on the operator’s proximity, following the guidelines of ISO10218-2 [5]. A look-ahead mechanism predicts human movements over a time window and adapts robot behavior in advance, ensuring that transitions in human tasks do not create unsafe situations. The same mechanism is replicated in the simulation environment, to account for safety slowdowns in the selection of the optimal schedule and layout. By combining packing layout, base pose optimization, task sequencing, and human-aware adaptation, the framework provides an integrated solution for optimizing the behavior of mobile manipulators in collaborative pick-and-place scenarios. The result is a balance between cycle time, efficiency and compliance with human safety requirements. III. EXPERIMENTS The proposed framework was experimentally validated using an ABB CRB 15000 collaborative robot mounted on a linear axis, enabling the search for optimal base positions along a 3-meter range. The case study in shown in Fig. 2b. The experiments were performed in a planar box-packing scenario with two object types and a single box available for each type. Pick-and-place operations were executed by combining predefined motion primitives (visible in the accompanying video), while the human operator carried out tasks such as preparing boxes, transporting items, and filling pallets (we assumed that the human could only be in one of the four colored areas in Fig. 2b). To support optimization, an inference dataset was created by running thousands of randomized pick-and-place trials. Execution time and robot manipulability were collected under nominal conditions to provide baseline statistics for normalizing the optimization objectives. This approach allowed the system to account for variations in task durations and the impact of safety-related velocity reductions when the human operator was near the workspace. An example of the schedule generated by the proposed optimizer is shown in Fig. 2a, together with a visual representation of the lookahead approach we implemented. The results show that the Particle Swarm Optimization framework effectively balances productivity and safety. Compared to a randomized baseline, the proposed method achieved a reduction in the cycle time of around 12% with significantly less computational effort. When task priorities were modified, the method successfully adapted the search, prioritizing other metrics such as ergonomics or human-robot distance maximization. Based on this, our framework proved to be capable of handling scenarios characterized by different operating conditions, both in terms of human schedule and in terms of target KPIs. Additionally, scenarios without human presence were tested, to show the flexibility of the pipeline. The cycle time reduction was around 8%, consistently with the results obtained in the collaborative case. Overall, the experiments validate that the framework improves task efficiency, adapts to different human task patterns, and maintains safety in collaborative pick-and-place operations. IV. CONCLUSION We presented a framework for optimizing mobile manipulators in human-robot collaborative tasks. By jointly determining robot base positions and task sequencing while accounting for human safety, the method improves efficiency and adaptability. Experiments demonstrate reduced cycle times across different human tasks. REFERENCES [1] C. Cella, M. B. Robin, M. Faroni, A. M. Zanchettin, and P. Rocco, “Digital model-driven genetic algorithm for optimizing layout and task allocation in human-robot collaborative assemblies,” IEEE ICRA, 2025. [2] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in IEEE International Conference on Neural Networks, 1995, pp. 1942–1948. [3] C. Cella, S. Sonnino, M. Faroni, A. M. Zanchettin, and P. Rocco, “Optimized scheduling and positioning of mobile manipulators in collaborative applications,” in IFAC Joint Conference on Computers, Cognition and Communication, 2025, pp. 1–6. [4] E. Dube, L. R. Kanavathy, and P. Woodview, “Optimizing threedimensional bin packing through simulation,” in Sixth IASTED International Conference Modelling, Simulation, and Optimization, 2006. [5] “ISO 10218-2:2025 Robots and robotic devices — Safety requirements for industrial robots,” International Organization for Standardization, Geneva, CH, Standard, 2025. 164