Force-Safe Environment Maps and Real-Time Detection for Soft Robot Manipulators

Akua K. Dickson, Juan C. Pacheco Garcia, Andrew P. Sabelhaus · 2025

Read Force-Safe Environment Maps and Real-Time Detection for Soft Robot Manipulators by Akua K. Dickson, Juan C. Pacheco Garcia, Andrew P. Sabelhaus (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-05307.txt.

Introduction

Akua Dickson and co-authors build a framework that keeps soft robot arms within safe force limits when they touch fragile things near them. Soft arms suit delicate work, but past obstacle methods track position only and ignore how hard the arm may push.

Their maps turn force safety rules from task space, points along the arm, into configuration space, the arm's joint angles. That lets the robot check force safety in real time as it moves. The work aims at safe use in homes, labs, and care sites where a hard push could do harm.

Catalog record

CREATORAkua K. Dickson, Juan C. Pacheco Garcia, Andrew P. Sabelhaus
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-05307

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Force-Safe Environment Maps and Real-Time Detection for Soft Robot Manipulators Akua K. Dickson, Juan C. Pacheco Garcia, Andrew P. Sabelhaus Abstract Soft robot manipulators have the potential for deployment in delicate environments to perform complex manipulation tasks. However, existing obstacle detection and avoidance methods do not consider limits on the forces that manipulators may exert upon contact with delicate obstacles. This work introduces a framework that maps force safety criteria from task space (i.e. positions along the robot’s body) to configuration space (i.e. the robot’s joint angles) and enables real-time force safety detection. We incorporate limits on allowable environmental contact forces for given task-space obstacles, and map them into configuration space (C-space) through the manipulator’s forward kinematics. This formulation ensures that configurations classified as safe are provably below the maximum force thresholds, thereby allowing us to determine force-safe configurations of the soft robot manipulator in real-time. We validate our approach in simulation and hardware experiments on a two-segment pneumatic soft robot manipulator. Results demonstrate that the proposed method accurately detects force safety during interactions with deformable obstacles, thereby laying the foundation for real-time safe planning of soft manipulators in delicate, cluttered environments. I Introduction Soft robot manipulators are widely known for their ability to interact compliantly with their environments to perform delicate tasks such as medical procedures [ 1 , 2 ] . Their intrinsic compliance often results in lower contact forces compared to rigid counterparts [ 3 ] , which has led to the common perception that soft robots are inherently safe. However, compliance alone may not guarantee safety. The magnitude of environmental contact forces during robot motion depends not only on the robot’s mechanical properties but also on its dynamics and the stiffness of the environment [ 4 ] . For sensitive tasks such as medical procedures, it is desirable to limit these environmental contact forces below certain limits [ 2 ] . This motivates a shift from purely geometric collision detection and avoidance toward force safety, where safety is defined by limits on contact forces rather than by the absence of contact. Collision detection and obstacle avoidance have been extensively studied for rigid robot manipulators [ 5 , 6 ] . Many task-space planning algorithms neglect the robot’s full-body dynamics and often produce trajectories that are not dynamically feasible, particularly for systems with high degrees of freedom (DOF) [ 7 , 8 ] . To address these limitations, motion planning is reformulated in the configuration space [ 9 , 10 , 11 ] , where each point represents a unique robot configuration. This transformation enables efficient safe motion planning using sampling based approaches such as probabilistic roadmaps (PRM) [ 12 , 13 ] , rapidly exploring random trees (RRTs) [ 14 , 15 ] and optimization-based approaches [ 16 , 17 ] . While effective for collision-free motion planning of rigid robot manipulators, these methods do not directly extend to compliant soft manipulators. Furthermore, traditional collision detection treats safety as a purely geometric notion, i.e., the manipulator is considered safe as long as its body does not intersect any task space obstacle [ 18 , 19 ] . As such, these approaches do not consider obstacle deformation and force safety criteria (Defn. 1 below) [ 20 ] . Fig. 1: Soft robot manipulator motions in hardware and simulation are overlaid on top of each other, with colors indicating force safety: green for safe and red for unsafe configurations. (a) shows a timelapse of motion toward the first obstacle, and (b) toward the second obstacle. In contrast, soft robot manipulators possess infinite or very high DOF, making both configuration space representation and collision reasoning far more challenging. Existing work in soft robot motion planning often extends rigid-robot sampling-based methods such as RRT* [ 21 ] and potential field methods [ 22 ] under kinematic assumptions such as the piecewise constant-curvature approximation. Furthermore, task-space collision avoidance methods [ 23 ] generate collision-free trajectories while considering kinematic constraints of soft manipulators. Configuration-space based collision detection and motion planning methods [ 24 , 25 ] on the other hand, further enhance accuracy and dynamic feasibility by leveraging the exact constant-curvature geometry and operating within the state space of the robot. While these methods enable collision detection and obstacle avoidance, they typically treat collisions as strictly undesirable. As a result, they do not provide guarantees on force safety or account for environmental contact forces. This manuscript bridges this gap by presenting a novel approach for real-time force-unsafe deformation detection for soft robot manipulators (Fig. 1 ). We contribute: • A method (Fig. 2 ) to map force safety criteria for soft robot manipulators into the robot’s configuration space. • A method to detect force-safe states in real time. • A validation in both simulation and hardware, demonstrating that our method detects force safety. Fig. 2: Overview of our force-safe environment maps and real-time detection approach. (1) illustrates the deformation of a polygonal environment along its inward normal direction to define the force-unsafe deformation region (2). (3) shows inflation of the FODR obstacles by the manipulator’s thickness, allowing us to consider only the backbone for subsequent computations. (4) demonstrates our method for mapping FODR poses to configuration space and (5) shows computation of alpha shape regions for efficient force safety queries for multiple robot configurations. II Force-Unsafe Obstacle Maps in Task Space We adopt the notion of force safety as introduced in [ 20 ] , where safety is not defined by zero collisions with the environment but by environmental collisions that are bounded by maximum force thresholds. Definition 1 Force Safety: Following [ 20 ] , a soft robot manipulator’s motion is said to be force-safe if its applied contact forces \mathbf{F}(t) remain within a known admissible set F_{\mathrm{safe}} for all time, i.e., \mathbf{F}(0)\in\mathcal{F}_{\mathrm{safe}}\;\;\Rightarrow\;\;\mathbf{F}(t)\in\mathcal{F}_{\mathrm{safe}},\quad\forall t\geq 0. Equivalently, force safety is achieved when \mathcal{F}_{\mathrm{safe}} is an invariant set of the system dynamics. To connect geometry with force limits, we adopt a model of the deformable obstacle environment as a polytope \mathcal{N}=\{\mathbf{r}\mid H\mathbf{r}\leq\mathbf{h}\} representing the original obstacle region (Fig. 3 ). Each facet of this polytope is assumed to deform elastically in its inward normal direction, with a force–displacement relationship \psi(\cdot) . A maximum admissible contact force F^{max} then corresponds to a maximum deflection n^{max}=\delta^{-1}\psi^{-1}(F^{max}) , where \delta\in[0,1] is a safety factor. Translating each facet of \mathcal{N} inward by n^{max} yields a new polytope \mathcal{P}=\{\mathbf{r}\mid H^{\prime}\mathbf{r}\leq\mathbf{h}^{\prime}\}, which represents the set of robot poses that are guaranteed to apply more than F^{max} to the obstacles in the environment. Here, H^{\prime}=H , and \mathbf{h}^{\prime}=\mathbf{h}-n^{max}\sqrt{m^{2}+1} where m is the slope of each hyperplane that forms the polytope \mathcal{P} . As shown in [ 20 ] , this translation provides a conservative but tractable approximation of force safety in task space. The set \mathcal{P} can be interpreted as a force-unsafe obstacle deformation region in task space as seen in Fig. 3 . This construction encodes both geometry and allowable contact forces. Therefore, the boundary of \mathcal{P} distinguishes between robot poses that are (i) collision-free, (ii) in contact but force-safe, and (iii) unsafe due to exceeding F^{max} . This task-space representation forms the foundation for our subsequent mapping into configuration space, where we construct force-safe configuration space obstacle maps

Read the complete text