Robot Graph Grammars: Towards Custom Robots for Every Task

As robots find broader applications outside factory floors, they face an increasing number of challenges. For example, they must accommodate rugged terrain, limited battery capacity, and complex dynamics. Existing robots are largely designed by hand to meet a given set of specifications. While highly capable, these manually-designed robots tend to leave performance on the table. These difficulties have motivated research into automatic robot design tools. Early tools were often limited in the range of robot topologies they can explore, however. Current graph-based robot representations can expand the space of possible designs, but it is not always clear how the resulting designs can be fabricated. To enable efficient design exploration and ensure fabricability, we propose graph grammars as a universal robot design representation. Graph grammars use rewriting rules to incrementally add complexity or select among distinct design alternatives. Because only fabricable components and connections are expressed in the grammar, the generated robot topologies are valid by construction. Through recursion and branching, graph grammars can also generate a large variety of possible designs. To tackle this expansive search space, we propose a specialized learning-based search algorithm called Graph Heuristic Search (GHS). GHS focuses limited simulation resources on the most promising designs. We compare GHS to random search and Monte-Carlo tree search baselines, showing that GHS finds higher-performing designs in less wall-clock time. We combine graph grammars and GHS with other techniques such as differentiable simulation to efficiently optimize multiple types of mobile robots. In doing so, we show that graph grammars are a principled yet general design representation for robot co-design. Their efficiency and versatility brings us one step closer to the dream of generating custom robots for every task.

Name thesis:degree_name

Doctoral

Department dc:contributor.department

Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2024

Author dc:creator

• Zhao, Allan

Advisors dc:contributor.advisor

• Matusik, Wojciech
• Rus, Daniela

Statement dc:rights

• In Copyright - Educational Use Permitted
• Copyright retained by author(s)

Licence dc:rights.uri

• https://rightsstatements.org/page/InC-EDU/1.0/

Handle dc:identifier.uri

https://hdl.handle.net/1721.1/153850

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/153850