Scavenger: Real-time logic-based control for an autonomous scavenger robot

Objective
This project aims to develop an open-source ROS-compatible real-time logic-based integrated planning, reasoning and control system for mobile robots. The key novelty in our project is including non-axiomatic reasoning in the robot software stack to complement common techniques such as deep learning in handling uncertainty. The system will be featured in a scavenger — a mobile robot used to inspect a city-like environment to carry out a collection of pieces of waste. With the final demonstrator, we aim to showcase the potential of our integrated planning, reasoning, and control system for mobile robots that need to carry out tasks in unknown environments.

Background
Today’s robotic control systems rely on big data, machine learning approaches and/or extensive (physical) modelling and behaviour pre-programming to achieve their required functionalities. While still utilizing such techniques, this demonstrator aims to introduce improvements towards low-energy, cost-efficient and effective mobile robots by integrating a reasoning-based system, the Non-Axiomatic Reasoning System (NARS). NARS is designed to build mission-relevant hypotheses from a stream of input events and to act upon the most successful predicting hypotheses. With its ability to learn and update hypotheses in real-time with little training or task pre-programming, NARS will be the key technology allowing our robot to improvise in challenging and uncertain situations, identify new types of objects and categorize them based on their perceived properties.

Crossdisciplinary collaboration
The researchers in the team represent the KTH School of Electrical Engineering and Computer Science, Division of Robotics, Perception and Learning and KTH School of Industrial Engineering and Management, Department of Machine Design.

Watch the recorded presentation at the Digitalize in Stockholm 2023 event: