Project Abstract

High-speed delivery and transfer of goods and materials requires fast, effective, and reliable logistics systems both inside warehouses as well as in the transportation system. Getting this right is especially crucial in our highly fragmented globalized supply chains. In our comprehensive research suite for smart logistics, we are developing tools and automation technologies to enhance the capabilities of modern warehouses.

We delved into three pivotal realms: real-time human attention estimation, human motion predictability, and symbolic regression enhancing interpretability. Through innovative methodologies and metrics, our works revolutionize human-robot collaborations and transparent prediction frameworks. These breakthroughs collectively promise to accelerate next-generation logistics operations with significantly improved safety and efficiency.

Project Results

Attention in Perception:

In this work, we introduced a methodology for quantifying the relationship between perception metrics and robot metrics, taking into account factors such as detection rate, detection quality, and latency. Furthermore, we introduced two novel metrics for Human-Robot Collaboration safety predicated upon perception metrics: Critical Collision Probability (CCP) and Average Collision Probability (ACP). To validate the utility of these metrics in facilitating algorithm development and tuning, we developed an attentive processing strategy that focuses exclusively on key input features. This approach significantly reduces computational time while preserving a similar level of accuracy.

Human Motion Prediction:

Human motion prediction is an essential step for efficient and safe human-robot collaboration. Current methods are missing out on leveraging well-studied human body kinematic models as well as body and scene constraints which can help boost the efficacy of these prediction frameworks while also explicitly avoiding implausible human joint configurations. We proposed a novel human motion prediction framework that incorporates human joint constraints and scene constraints in a Gaussian Process Regression (GPR) model to predict human motion over a set time horizon. This formulation is combined with an online context-aware constraints model to leverage task-dependent motions. It is tested on a human arm kinematic model and implemented on a human-robot collaborative setup with a UR5 robot arm to demonstrate the real-time capability of our approach. Simulations were also performed on datasets like HA4M and ANDY. The simulation and experimental results demonstrate considerable improvements in a Gaussian Process framework when these constraints are explicitly considered.

Symbolic Regression and Interpretability:

In contrast to neural networks which are often treated as black boxes, Symbolic Regression (SR) identifies models explicitly represented as mathematical (symbolic) expressions. We proposed a Generalized Symbolic Regression (GSR) task, which generalizes the classical SR task of identifying expressions of the form y = f(x), by discovering analytical mappings from the input space to a transformed output space through expressions of the form g(y) = f(x). Our proposed approach achieves promising experimental performance, highlighting new insights, which can open up new research directions in SR.

Further, we showed how SR can be leveraged in recovering probability density functions from samples via Maximum Entropy Distribution (MED) ansatz. We find that:

1. i) The Lagrange multipliers of MED can be simply found by solving a linear system of equations, eliminating the need for computationally expensive Newton’s method.
2. ii) Accurate density recovery does not necessarily require relying on higher-order moments. In fact, increasing the number of complex but low-order symbolic basis functions leads to better conditioning and enhanced expressiveness.

KACST PI: Ahmad Alabdulkareem

MIT PI: Kamal Youcef-Toumi