Autonomous Manipulation of Hazardous Chemicals and Delicate Objects in a Self-Driving Laboratory: A Sliding Mode Approach

Precise handling of chemical instruments and materials within a self-driving laboratory environment using robotic systems demands advanced and reliable control strategies. Sliding Mode Control (SMC) has emerged as a robust approach for managing uncertainties and disturbances in manipulator dynamics, providing superior control performance compared to traditional methods. This study implements a model-based SMC (MBSMC) utilizing a hyperbolic tangent function to regulate the motion of a manipulator mounted on a mobile platform operating inside a self-driving chemical laboratory. Given the manipulator’s role in transporting fragile glass vessels filled with hazardous chemicals, the controller is specifically designed to minimize abrupt transitions and achieve gentle, accurate trajectory tracking. The proposed controller is benchmarked against a non-model-based SMC (NMBSMC) and a Proportional-Integral-Derivative (PID) controller using a comprehensive set of joint and Cartesian metrics. Compared to PID and NMBSMC, MBSMC achieved significantly smoother motion and up to 90% lower control effort, validating its robustness and precision for autonomous laboratory operations. Experimental trials confirmed successful execution of tasks such as vessel grasping and window operation, which failed under PID control due to its limited ability to handle nonlinear dynamics and external disturbances, resulting in substantial trajectory tracking errors. The results validate the controller’s effectiveness in achieving smooth, precise, and safe manipulator motions, supporting the advancement of intelligent mobile manipulators in autonomous laboratory environments.

💡 Research Summary

The paper addresses the challenge of safely handling fragile glass vessels containing hazardous chemicals within a self‑driving laboratory by introducing a model‑based sliding mode controller (MBSMC) for a mobile manipulator. The hardware platform consists of a Ridgeback omnidirectional base carrying a Universal Robots UR5e 6‑DOF collaborative arm equipped with a Robotiq Hand‑E adaptive gripper, supplemented by LIDAR and RealSense vision for localization and obstacle avoidance. Traditional PID control, while widely used, suffers from poor disturbance rejection and abrupt torque commands that can break delicate labware. To overcome these limitations, the authors develop an SMC scheme that explicitly incorporates the full robot dynamics (inertia, Coriolis‑centrifugal, and gravity terms) and defines a sliding surface σ(x)=P₁e+P₂ẋ. The control input is split into an equivalent control term and a switching term. Crucially, the switching term uses a hyperbolic tangent (tanh) function instead of the conventional sign function, dramatically reducing high‑frequency chattering while preserving the robustness characteristic of sliding mode control. Tuning parameters P₁, P₂ govern convergence speed and robustness, whereas P₃ adjusts the rate of transition on the sliding surface.

For benchmarking, the MBSMC is compared against a first‑order non‑model‑based SMC (NMBSMC) and a classic PID controller. Evaluation metrics include joint‑space and Cartesian position errors, velocity, acceleration, jerk, snap, and control‑effort measured as RMS torque. Simulations in Gazebo introduce external disturbances such as wind gusts and base slip, demonstrating that MBSMC maintains the smallest tracking error and the smoothest torque profile. Experimental validation involves three representative laboratory tasks: (1) grasping and transporting glass vials, (2) operating an automated electrolyte‑mixing system, and (3) opening a laboratory window. Under PID control, the manipulator exhibits large torque spikes that cause glass breakage and failure to fully open the window. NMBSMC reduces the error but still generates noticeable chattering, leading to excessive jerk and potential damage to sensitive samples. In contrast, MBSMC successfully completes all tasks with an average position error below 0.02 rad, a 90 % reduction in control effort relative to PID, and an 85 % reduction compared to NMBSMC. High‑order smoothness metrics (jerk and snap) confirm the gentle motion required for fragile objects.

The contributions of the work are threefold: (1) a novel hyperbolic‑tangent‑based sliding mode controller that suppresses chattering while retaining robustness, (2) the introduction of jerk and snap as quantitative smoothness indicators for laboratory‑grade manipulation, and (3) a comprehensive validation pipeline that spans physics‑based simulation and real‑world experiments in a chemical laboratory setting. Limitations include the need for manual retuning of the controller gains when operating conditions change and the absence of an explicit compensation mechanism for base slip during high‑speed navigation. Future research directions propose integrating adaptive gain estimation and slip‑observer modules into the SMC framework, as well as extending the approach to multi‑robot collaborative scenarios to further advance autonomous laboratory automation.