A Robot-Agnostic Kinematic Control Framework: Task Composition via Motion Statecharts and Linear Model Predictive Control

Abstract

Autonomous service robots frequently struggle to execute motion tasks that high- level planners describe, especially when these tasks involve varying robots, envi- ronments, and object types. This thesis presents a framework for kinematic robot motion execution that addresses this motion execution gap by focusing on five core properties: transferability, composability, reactivity, whole-body control, and smooth motion task transitions. The proposed approach employs a world-centric kinematic model that includes both the robot and its environment, enabling consistent definitions of common actions (such as door or drawer opening) across different platforms. Motion tasks and constraints are expressed in Motion Statecharts, which structure sequential and parallel behaviors, allowing composition of complex motion sequences while maintaining real-time reactivity. A key innovation lies in extending a task function approach controller with linear model predictive control, thereby constraining jerk and acceleration to ensure robust, smooth execution for Motion Statecharts. This thesis evaluates the system on nine robots, ranging from mobile manip- ulators to stationary dual-arm setups, and in diverse scenarios, including retail environments and domestic-service applications. Empirical results demonstrate that the same parameterization and core motion task definitions can be transferred across different kinematics and environments with minimal modifications. Further experiments highlight how the model predictive control formulation integrates smoothly with monitors for collision avoidance, in-hand manipulation, and visual servoing. Overall, the developed framework unifies composable statechart-based motions, real-time reactivity, and smooth multi-degree-of-freedom control, constituting a practical solution for bridging the motion execution gap in modern service robotics.