Reactive Motion Planning and Control under Constraints
Doctoral thesis, 2023
Modern robots are increasingly being designed to operate in dynamic and unstructured environments shared with humans and other mobile agents. In these scenarios, the robot must react to any online detected changes to provide a correct and safe operation.
This thesis introduces techniques for motion planning and control employing Dynamical Systems (DSs) formulations, such as Artificial Potential Fields (APF) and Dynamical Movement Primitives (DMP). While several DS-based methods facilitate reactivity to changes in the robot environment with guaranteed convergence to a specified goal, they often lack the integration of robot constraints. Moreover, convergence depends on specific environmental conditions, e.g., the environment being a star world.
To extend the range of practical scenarios in which convergence to the goal is guaranteed in star worlds, a method is proposed for online adjustment of the robot's perceived environment to align with the necessary conditions. This method is enabled through a comprehensive exploration of the concept starshaped hull. To account for robot constraints, two strategies are proposed. The first approach employs online modification of the DS through a scaling factor to meet velocity and acceleration constraints. Compared to standard scaling approaches, focus is placed on mitigating the feasibility issues that are more prominent in reactive motion planning. In particular, by proactively scaling the DS before reaching the acceleration limits, the proposed method retains the feasibility for a wider set of trajectories. The efficacy of the scaling is showcased by experiments on a robotic arm. The second approach integrates a DS method within a Model Predictive Control (MPC) scheme. This integration effectively combines the favorable convergence characteristics of the DS with the intuitive representation of system constraints offered by MPC. In this configuration, the DS generates a path with an obstacle clearance, while the MPC provides a feasible trajectory that adheres to this clearance distance. The scheme is proven to accomplish collision avoidance and to preserve the convergence guarantees provided by the DS for a substantial portion of the workspace. Furthermore, the scheme is extended to encompass path-following control as well.
This thesis introduces techniques for motion planning and control employing Dynamical Systems (DSs) formulations, such as Artificial Potential Fields (APF) and Dynamical Movement Primitives (DMP). While several DS-based methods facilitate reactivity to changes in the robot environment with guaranteed convergence to a specified goal, they often lack the integration of robot constraints. Moreover, convergence depends on specific environmental conditions, e.g., the environment being a star world.
To extend the range of practical scenarios in which convergence to the goal is guaranteed in star worlds, a method is proposed for online adjustment of the robot's perceived environment to align with the necessary conditions. This method is enabled through a comprehensive exploration of the concept starshaped hull. To account for robot constraints, two strategies are proposed. The first approach employs online modification of the DS through a scaling factor to meet velocity and acceleration constraints. Compared to standard scaling approaches, focus is placed on mitigating the feasibility issues that are more prominent in reactive motion planning. In particular, by proactively scaling the DS before reaching the acceleration limits, the proposed method retains the feasibility for a wider set of trajectories. The efficacy of the scaling is showcased by experiments on a robotic arm. The second approach integrates a DS method within a Model Predictive Control (MPC) scheme. This integration effectively combines the favorable convergence characteristics of the DS with the intuitive representation of system constraints offered by MPC. In this configuration, the DS generates a path with an obstacle clearance, while the MPC provides a feasible trajectory that adheres to this clearance distance. The scheme is proven to accomplish collision avoidance and to preserve the convergence guarantees provided by the DS for a substantial portion of the workspace. Furthermore, the scheme is extended to encompass path-following control as well.
computational geometry
online trajectory generation
navigation
optimization-based control
robotics
dynamical systems
SB-H4, Sven Hultins Gata 6
Opponent: Associate Professor Dimitra Panagou, Department of Robotics and Department of Aerospace Engineering, University of Michigan, USA.
Author
Albin Dahlin
Chalmers, Electrical Engineering, Systems and control
Adaptive Trajectory Generation under Velocity Constraints using Dynamical Movement Primitives
IEEE Control Systems Letters,;Vol. 4(2020)p. 438-443
Journal article
Temporal Coupling of Dynamical Movement Primitives for Constrained Velocities and Accelerations
IEEE Robotics and Automation Letters,;Vol. 6(2021)p. 2233-2239
Journal article
Trajectory Scaling for Reactive Motion Planning
Proceedings - IEEE International Conference on Robotics and Automation,;(2022)p. 5242-5248
Paper in proceeding
Creating Star Worlds: Reshaping the Robot Workspace for Online Motion Planning
IEEE Transactions on Robotics,;Vol. In Press(2023)
Journal article
Dahlin, A., Karayiannidis, Y., Obstacle Avoidance in Dynamic Environments via Tunnel-following MPC with Adaptive Guiding Vector Fields
Dahlin, A., Karayiannidis, Y., Autonomous Navigation with Convergence Guarantees in Complex Dynamic Environments
Robots have historically operated in safety cages, where they have executed repetitive and pre-defined tasks. The robot motion can in such case be planned offline to optimize some performance criteria. Recent years have however witnessed a significant trend in ``liberating'' robots from their cages and many modern robots are specifically designed for operation in dynamically changing environments where they share space with humans and other mobile agents. In such situations, the robot cannot rely on a precomputed motion plan, but must react to any online detected changes to ensure a reliable and safe operation. The planning and execution of such responsive motions should also consider the constraints of the robot, including factors like acceleration limits and restrictions in motion directions.
This thesis explores the application of Dynamical System (DS) methods for reactive motion planning. While these techniques offer efficient means to react to changes in the robot's surroundings, addressing robot constraints is not a straightforward process. Moreover, ensuring movement to a specified goal position depends on particular environmental conditions, such as non-overlapping obstacles. In many real-world situations, closely positioned obstacles are however perceived by the robot as overlapping.
To address robot constraints, this thesis introduces two strategies: the first involves online temporal adjustment of the DS, and the second incorporates the DS in a constrained control problem. Additionally, an algorithm to adjust the robots perceived environment is proposed to allow for intersecting obstacles, thereby expanding the range of practical scenarios where progression towards the goal is assured.
This thesis explores the application of Dynamical System (DS) methods for reactive motion planning. While these techniques offer efficient means to react to changes in the robot's surroundings, addressing robot constraints is not a straightforward process. Moreover, ensuring movement to a specified goal position depends on particular environmental conditions, such as non-overlapping obstacles. In many real-world situations, closely positioned obstacles are however perceived by the robot as overlapping.
To address robot constraints, this thesis introduces two strategies: the first involves online temporal adjustment of the DS, and the second incorporates the DS in a constrained control problem. Additionally, an algorithm to adjust the robots perceived environment is proposed to allow for intersecting obstacles, thereby expanding the range of practical scenarios where progression towards the goal is assured.
UNICORN - Sustainable, Peaceful and Efficient Robotic Refuse Handling
VINNOVA (2017-03055), 2017-10-25 -- 2020-08-31.
Volvo Group, 2021-03-24 -- 2021-05-31.
Areas of Advance
Information and Communication Technology
Subject Categories
Robotics
ISBN
978-91-7905-972-9
Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 5438
Publisher
Chalmers
SB-H4, Sven Hultins Gata 6
Opponent: Associate Professor Dimitra Panagou, Department of Robotics and Department of Aerospace Engineering, University of Michigan, USA.