Incremental Semantically Grounded Learning from Demonstration

• DOI: 10.15607/rss.2013.ix.048
• Publication Date: 2016-01-03T02:49:07Z
• Authors: Scott Niekum; Sachin Chitta; Andrew Barto; Bhaskara Marthi; Sarah Osentoski
• Keywords: 0209 industrial biotechnology; 02 engineering and technology

Much recent work in robot learning from demonstration has focused on automatically segmenting continuous task demonstrations into simpler, reusable primitives.However, strong assumptions are often made about how these primitives can be sequenced, limiting the potential for data reuse.We introduce a novel method discovering semantically grounded and incrementally building improving finite-state representation of which various contingencies arise.Specifically, Beta Process Autoregressive Hidden Markov Model is used to segment motion categories, then further subdivided states automaton.During replay task, data-driven approach collect additional where they most needed through interactive corrections, improve automaton.Together, this allows intelligent sequencing create novel, adaptive behavior that improved as needed.We demonstrate utility technique furniture assembly using PR2 mobile manipulator.