Admittance-Based Motion Planning with Vision-Guided Initialization for Robotic Manipulators in Self-Driving Laboratories

TL;DR

Self-driving laboratories require compliant, force-aware manipulation for safe human-robot collaboration. This paper presents a motion-planning framework that embeds an admittance controller into trajectory execution and uses vision-based initialization to set the initial target pose, enabling dynamic responsiveness to external interaction. The admittance model treats the end-effector as a virtual mass–spring–damper with $M_d$, $B_d$, and $K_d$, converting external forces into a Cartesian reference velocity and blending it into the commanded trajectory $x_{cmd}$. Validation includes both simulation and real-world experiments with a textured object, showing compliant behavior under forces up to $18$ N and sub-centimeter pose accuracy in vision, underscoring improvements in safety, autonomy, and human–robot collaboration for SDLs; future work targets transparent/deformable objects and multi-arm scalability.

Abstract

Self driving laboratories (SDLs) are highly automated research environments that leverage advanced technologies to conduct experiments and analyze data with minimal human involvement. These environments often involve delicate laboratory equipment, unpredictable environmental interactions, and occasional human intervention, making compliant and force aware control essential for ensuring safety, adaptability, and reliability. This paper introduces a motion-planning framework centered on admittance control to enable adaptive and compliant robotic manipulation. Unlike conventional schemes, the proposed approach integrates an admittance controller directly into trajectory execution, allowing the manipulator to dynamically respond to external forces during interaction. This capability enables human operators to override or redirect the robot's motion in real time. A vision algorithm based on structured planar pose estimation is employed to detect and localize textured planar objects through feature extraction, homography estimation, and depth fusion, thereby providing an initial target configuration for motion planning. The vision based initialization establishes the reference trajectory, while the embedded admittance controller ensures that trajectory execution remains safe, adaptive, and responsive to external forces or human intervention. The proposed strategy is validated using textured image detection as a proof of concept. Future work will extend the framework to SDL environments involving transparent laboratory objects where compliant motion planning can further enhance autonomy, safety, and human-robot collaboration.

Figures (10)

• Figure 1: Lab environment with a mobile manipulator, a fixed manipulator, and other laboratory equipments
• Figure 2: Admittance control scheme for the manipulator
• Figure 3: Simulation setup for vision-guided grasping
• Figure 4: Output of vision algorithm (a)Boundary box (b)Axes of frames
• Figure 5: Motion of manipulator to grasp a book during simulation