- Title Information
- Title
- Robotic Language Grounding
- Type of Resource (primo)
- dissertations
- Name:
Personal
- Name Part
- Liu, Jason Xinyu
- Role
- Role Term:
Text
- creator
- Name:
Personal
- Name Part
- Tellex, Stefanie
- Role
- Role Term:
Text
- Advisor
- Name:
Personal
- Name Part
- Konidaris, George
- Role
- Role Term:
Text
- Reader
- Name:
Personal
- Name Part
- Littman, Michael
- Role
- Role Term:
Text
- Reader
- Name:
Personal
- Name Part
- Pavlick, Ellie
- Role
- Role Term:
Text
- Reader
- Name:
Personal
- Name Part
- Shah, Julie
- Role
- Role Term:
Text
- Reader
- Name:
Corporate
- Name Part
- Brown University. Department of Computer Science
- Role
- Role Term:
Text
- sponsor
- Origin Information
- Copyright Date
- 2025
- Physical Description
- Extent
- 19, 134 p.
- digitalOrigin
- born digital
- Note:
thesis
- Thesis (Ph. D.)--Brown University, 2025
- Genre (aat)
- theses
- Abstract
- Robots are becoming capable and prevalent in human environments. It is crucial to specify tasks accurately so that robot execution is effective and safe. Natural language provides an intuitive, expressive, and flexible way for humans to communicate with robots. With advances in large language models (LLMs) and vision-language models (VLMs), we should be able to talk to robots like talking with other people. However, situating diverse and complex language in the physical world is challenging. Grounding language to a structured task specification, like linear temporal logic (LTL), enables autonomous robots to understand a broad range of natural language and solve long-horizon tasks with safety guarantees, and its compositionality can induce skill transfer. 1) This thesis first introduces a framework for situating work on grounding language onto various representations and proposes desired properties to guide the discussion of their tradeoffs. Next, we propose two works on building modular language grounding systems. 2) Lang2LTL uses pretrained and finetuned LLMs to ground language with diverse temporal patterns to LTL task specifications in novel environments without retraining on language data . 3) Improved upon its predecessor, Lang2LTL-2 uses pretrained and finetuned LLMs and a pretrained VLM to ground language with diverse spatiotemporal constraints zero-shot.
By translating language to LTL, the robot can detect infeasible task specifications and abort execution when necessary. 4) Finally, this thesis introduces LTL-Transfer, a zero-shot transfer algorithm that leverages the compositionality of LTL to reuse learned skills to solve novel tasks without violating any safety constraints. Future work will focus on developing robotic systems that produce robust and verifiable robot behavior by integrating multimodal grounding and human-robot dialogue.
- Subject
- Topic
- human-robot interaction
- Subject (fast)
(authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/01098997")
- Topic
- Robotics
- Language
- Language Term (ISO639-2B)
- English
- Record Information
- Record Content Source (marcorg)
- RPB
- Record Creation Date
(encoding="iso8601")
- 20251201