[Paper Review] Grasp that optimises objectives along post-grasp trajectories

Amir M. Ghalamzan E., Nikos Mavrakis+1 more|arXiv (Cornell University)|Dec 12, 2017

Robot Manipulation and Learning14 references3 citations

TL;DR

This paper proposes a multi-objective grasp selection framework that optimizes kinematic manipulability, joint torque effort, and impact force along predefined post-grasp trajectories. By evaluating 10 grasp poses on a cuboid object across three Pick-and-Place tasks, the study demonstrates that these objectives often conflict, necessitating multi-objective optimization for optimal grasp selection.

ABSTRACT

In this article, we study the problem of selecting a grasping pose on the surface of an object to be manipulated by considering three post-grasp objectives. These objectives include (i) kinematic manipulation capability, (ii) torque effort \cite{mavrakis2016analysis} and (iii) impact force in case of a collision during post-grasp manipulative actions. In these works, the main assumption is that a manipulation task, i.e. trajectory of the centre of mass (CoM) of an object is given. In addition, inertial properties of the object to be manipulated is known. For example, a robot needs to pick an object located at point A and place it at point B by moving it along a given path. Therefore, the problem to be solved is to find an initial grasp pose that yields the maximum kinematic manipulation capability, minimum joint effort and effective mass along a given post-grasp trajectories. However, these objectives may conflict in some cases making it impossible to obtain the best values for all of them. We perform a series of experiments to show how different objectives change as the grasping pose on an object alters. The experimental results presented in this paper illustrate that these objectives are conflicting for some desired post-grasp trajectories. This indicates that a detailed multi-objective optimization is needed for properly addressing this problem in a future work.

Motivation & Objective

• To investigate whether post-grasp objectives—kinematic manipulability, torque effort, and impact force—conflict during manipulative tasks.
• To evaluate how grasp pose selection affects these objectives along predefined center-of-mass trajectories.
• To demonstrate that single-objective grasp selection may be suboptimal due to conflicting performance metrics.
• To provide empirical evidence supporting the need for multi-objective optimization in task-informed grasp selection.
• To enable robots to select grasps that balance manipulability, energy efficiency, and safety during post-grasp motion.

Proposed method

• The study uses a simulated Baxter robot in Gazebo to evaluate 10 grasp poses on a 0.5×0.15×0.2 m cuboid object with known mass and inertia tensor.
• Three Pick-and-Place tasks were defined, each with a predefined center-of-mass trajectory for the object.
• For each grasp and trajectory, the paper computes three key metrics: Task-Oriented Velocity Manipulability (TOV), Torque Effort Metric (TEM), and Total Mechanical Energy (TME) as a proxy for impact force.
• The metrics are integrated along the trajectory using the L2 norm, yielding scalar values per grasp pose for comparison.
• Normalised values of TOV, TEM, and TME are computed and visualized across tasks to compare performance across grasps.
• Heat maps and line plots (e.g., Fig. 5 and Fig. 6) illustrate how each metric varies with grasp pose and trajectory waypoint.

Experimental results

Research questions

• RQ1 Do the objectives of kinematic manipulability, torque effort, and impact force consistently rank the same grasp pose as optimal across different post-grasp trajectories?
• RQ2 How do the values of TOV, joint effort, and effective mass vary with different grasp poses on the same object?
• RQ3 To what extent do the three objectives conflict in real-world robotic manipulation tasks?
• RQ4 Can a single grasp pose optimally satisfy all three objectives simultaneously for a given trajectory?
• RQ5 Is multi-objective optimization necessary for effective grasp selection when multiple performance criteria are involved?

Key findings

• For Task 1, grasp number 1 achieved the highest manipulability, lowest joint effort, and lowest effective mass, indicating it was optimal across all metrics.
• In Task 2, the objectives conflicted: grasp number 1 minimized effort and effective mass, but grasp number 6 maximized manipulability.
• In Task 3, grasp number 2 minimized joint effort, while grasp number 1 was optimal for both manipulability and effective mass.
• The results show that no single grasp pose optimizes all three objectives simultaneously in multiple tasks, indicating inherent conflicts.
• The normalised metric plots (Fig. 6) clearly demonstrate that performance rankings vary significantly across grasps and tasks, validating the need for multi-objective optimization.
• The effective mass varied spatially across the object’s surface and along the trajectory, with peak values observed at specific waypoints depending on the grasp pose.