Advances in Robotic Manipulation: Generalization and Uncertainty Handling

The recent advancements in robotic manipulation research are significantly pushing the boundaries of what is possible in spatial grasping, contact-grasping, and manipulation planning. Innovations are being driven by a combination of simulation-based data generation, novel probabilistic modeling, and the integration of geometric and learning-based approaches. Notably, there is a strong emphasis on enhancing generalization capabilities, addressing uncertainty in grasping, and improving the adaptability of robotic systems to diverse and complex environments. These developments are paving the way for more robust and versatile robotic manipulation systems that can operate effectively in real-world scenarios with varying levels of noise and uncertainty. Particularly noteworthy are approaches that leverage simulation data for policy training, probabilistic models for grasp detection, and intrinsic metrics for manipulation planning, all of which contribute to the field's progress towards more intelligent and autonomous robotic systems.

Sources

ManiBox: Enhancing Spatial Grasping Generalization via Scalable Simulation Data Generation

Diffusion-based Virtual Fixtures

vMF-Contact: Uncertainty-aware Evidential Learning for Probabilistic Contact-grasp in Noisy Clutter

Problem Space Transformations for Generalisation in Behavioural Cloning

Planning for quasi-static manipulation tasks via an intrinsic haptic metric

SuperQ-GRASP: Superquadrics-based Grasp Pose Estimation on Larger Objects for Mobile-Manipulation

Repairing Neural Networks for Safety in Robotic Systems using Predictive Models

Built with on top of