Cooperative Grasping for Collective Object Transport in Constrained Environments
David Felipe Alvear Goyes, George Turkiyyah, Shinkyu Park
AI summary
Problem
Identifying feasible grasp configurations for two mobile manipulators to collaboratively transport objects through constrained environments is computationally expensive and highly sensitive to diverse object geometries and complex spatial constraints.
Approach
The framework uses two neural networks to project grasp configurations into a shared embedding space, trained via supervised learning and negative sampling to score and rank feasible robot grasp pairs.
Key results
- Reliably identifies feasible grasp configurations across diverse environments and object geometries in simulation
- Reduces computational complexity from quadratic exhaustive search to efficient embedding-based scoring
- Successfully validates cooperative transport on a physical two-robot mobile manipulator platform
- Leverages negative sampling to accelerate training and handle class imbalance between feasible and infeasible pairs
Why it matters
Enables scalable, reliable multi-robot manipulation in complex real-world settings like warehouses or disaster response where narrow passages and varied object shapes are common.
Abstract
We propose a novel framework for decision-making in cooperative grasping for two-robot object transport in con- strained environments. The core of the framework is a Condi- tional Embedding (CE) model consisting of two neural networks that map grasp configuration information into an embedding space. The resulting embedding vectors are then used to identify feasible grasp configurations that allow two robots to collab- oratively transport an object. To ensure generalizability across diverse environments and object geometries, the neural networks are trained on a dataset comprising a range of environment maps and object shapes. We employ a supervised learning approach with negative sampling to ensure that the learned embeddings effectively distinguish between feasible and infeasible grasp configurations. Evaluation results across a wide range of environments and objects in simulations demonstrate the model’s ability to reliably identify feasible grasp configurations. We further validate the framework through experiments on a physical robotic platform, confirming its practical applicability.