Analyzing the utility of a support pin in sequential robotic manipulation

Chao Cao, Weiwei Wan, Jia Pan, Kensuke Harada

TL;DR

This paper introduces a novel approach to robotic pick-and-place manipulation by incorporating a vertical support pin as an intermediate placement surface. This simple addition significantly enhances a robot's ability to reorient objects, leading to higher success rates and more robust manipulation strategies, especially for challenging object shapes.

What Problem Does This Paper Address?

Robots often need to manipulate objects in ways that are impossible with a single grasp due to environmental constraints or the object's geometry. This requires "pick-and-place regrasp" – a sequence of picking up, placing, and re-grasping an object to incrementally change its pose. Traditional methods for this rely on placing objects on flat surfaces, which severely limits the number of stable poses an object can achieve. This limitation restricts the robot's options for reorientation and can lead to task failures, especially for complex or non-convex objects. The paper aims to overcome this by exploring how a simple vertical pin can provide more stable intermediate placements, thereby increasing the robot's manipulation capabilities.

Methods

The core of the proposed method lies in leveraging a vertical support pin to create more diverse and stable intermediate object placements. The process involves:

  1. Stable Placement Computation: The system automatically identifies stable placements of an object leaning against the pin and a flat surface. This is achieved by considering combinations of object edges (touching the flat surface) and sampled points on the object's surface (touching the pin). Stability, collision, and friction constraints are rigorously checked.
  2. Force-Closure Grasp Generation: For each valid placement, the algorithm finds multiple force-closure grasps that the robot can execute.
  3. Regrasp Graph Construction: A two-layer graph is built. The first layer connects stable placements, and the second layer connects the grasps associated with these placements. This hierarchical structure helps manage computational complexity.
  4. Sequence Planning: A shortest path search is performed on the regrasp graph to find an optimal sequence of pick-and-place actions to achieve the desired object pose.

The novelty lies in the systematic computation of object placements on a pin-supported surface, which significantly expands the state space of possible intermediate poses compared to flat-surface-only approaches.

Datasets

The experiments were conducted using various non-convex mesh models of objects, including a "pot lid like object" (1t1p) and an "L" shaped object (l), among others. These models were used to evaluate the performance of the proposed method across different shapes and complexities. The paper does not introduce new benchmark datasets but rather uses existing object models for evaluation.

Key Insights and Ideas

  • The Power of a Simple Pin: A seemingly minor addition like a vertical pin dramatically increases the number of stable object placements, offering a richer set of options for robotic manipulation.
  • Bridging the Gap for Difficult Objects: The pin is particularly beneficial for non-convex or complex objects (like pot lids) that are hard to reorient using only flat surfaces. It provides crucial intermediate poses that would otherwise be unreachable.
  • Connectivity is Key: The increased number of stable placements directly translates to a more connected regrasp graph, which is essential for finding successful manipulation sequences.
  • Optimal Pin Length: There's an optimal range for the support pin's length (around 2-4 cm in experiments) that maximizes success rates. Too short or too long can hinder stable placement.
  • Grasp Density Trade-off: While increasing grasp density generally improves success rates, the benefit plateaus after a certain point, suggesting a sweet spot for efficiency.

Key Results

  • Increased Success Rates: Experiments showed a significant improvement in reorientation task success rates when using the support pin, especially for challenging objects like the pot lid shape (1t1p). For some objects, the success rate with the pin was nearly 100% in areas where flat surfaces failed.
  • Enhanced Regrasp Graph Connectivity: The pin-based placements led to a more interconnected regrasp graph, providing more manipulation options.
  • Assembly Task Feasibility: A challenging assembly task that was impossible with only flat surfaces became feasible with the support pin.
  • Sequence Length: While the pin placement generally leads to higher success rates, it can sometimes result in slightly longer regrasp sequences for difficult cases, as more steps might be needed to achieve the goal when the task is inherently complex.

Limitations and Future Work

  • Uncertainty: The current framework assumes perfect knowledge of object geometry and ignores real-world uncertainties like object material properties (friction, weight distribution), which can affect grasp quality and stability.
  • Computational Cost: While the hierarchical graph structure helps, the pin-based approach can sometimes lead to larger regrasp graphs, potentially increasing computation time for graph building and search.
  • Future Directions:
    • Learning Optimal Parameters: Developing methods to learn optimal pin lengths and other parameters based on object properties and task requirements.
  • Robustness to Uncertainty: Incorporating robustness criteria and torque constraints to handle real-world uncertainties and ensure more reliable manipulation.
  • Real-World Implementation: Extending the framework to operate on physical robots for complex manufacturing tasks.

Summary updated

Abstract

Pick-and-place regrasp is an important manipulation skill for a robot. It helps a robot accomplish tasks that cannot be achieved within a single grasp, due to constraints such as kinematics or collisions between the robot and the environment. Previous work on pick-and-place regrasp only leveraged flat surfaces for intermediate placements, and thus is limited in the capability to reorient an object. In this paper, we extend the reorientation capability of a pick-and-place regrasp by adding a vertical pin on the working surface and using it as the intermediate location for regrasping. In particular, our method automatically computes the stable placements of an object leaning against a vertical pin, finds several force-closure grasps, generates a graph of regrasp actions, and searches for the regrasp sequence. To compare the regrasping performance with and without using pins, we evaluate the success rate and the length of regrasp sequences while performing tasks on various models. Experiments on reorientation and assembly tasks validate the benefit of using support pins for regrasping.
BibTeX
@inproceedings{cao2016analyzing,
  author = {Chao Cao and Weiwei Wan and Jia Pan and Kensuke Harada},
  title = {{Analyzing the utility of a support pin in sequential robotic manipulation}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2016},
  url = {https://openreview.net/forum?id=EoEwoH3inc},
}
Figure