Faculty of Engineering

ACFR Seminar Series [https://d31hzlhk6di2h5.cloudfront.net/20260723/44/a0/ac/85/83de99ae1318da0ad...] [https://d31hzlhk6di2h5.cloudfront.net/20260723/90/62/a1/a0/e806283be685b4d06...] Towards General Embodied Reasoning

Presented by Zeyu Zhang (UC Berkeley) Date: Friday, 24 July 2026, at 2:00 pm AEST Venue: ACFR seminar area, J04 level 2 (Rose St Buildinghttps://url.au.m.mimecastprotect.com/s/HNKJCBNqjlCMGgDLxfMcNf2WyeD?domain=t.e2ma.net) Zoom ID: https://uni-sydney.zoom.us/s/87306457455https://url.au.m.mimecastprotect.com/s/Sw1TCD1vlpTn9Ro87CxfRfj35gq?domain=t.e2ma.net Abstract:

Building general-purpose embodied agents requires moving beyond specialized systems toward models that can understand, reason about, and interact with diverse physical environments. In this talk, Zeyu Zhang will present his recent research on general embodied reasoning, covering 3D scene understanding, long-horizon navigation, robotic control, domain adaptation, and self-evolving agents. The talk will discuss how multimodal pretraining, scalable synthetic data, and data-centric learning can produce transferable representations that support a wide range of tasks, including captioning, grounding, question answering, dialogue, spatial reasoning, and planning. It will then examine how environmental understanding can be extended to embodied decision-making through hierarchical frameworks that combine fast reactive control with slower long-horizon reasoning. Zeyu will also discuss how carefully designed post-training, parameter-efficient adaptation, and verifiable feedback can improve navigation and control while preserving previously acquired perception and reasoning capabilities. Finally, the talk will explore how embodied agents can continue improving when manually specified objectives are incomplete or inaccurate, using self-supervised feedback, world-model-based exploration, adaptive memory, and stage-aware learning signals. Together, these studies highlight the importance of scalable data, spatially grounded foundation models, restrained reinforcement learning, and continual self-improvement for advancing toward general embodied intelligence.

Bio:

Zeyu is an incoming PhD student at UC Berkeley BAIR, advised by Professor Pieter Abbeel, Professor Alexei Efros, and Professor Angjoo Kanazawa. He received his bachelor’s degree from the Australian National University, where he was advised by Professor Richard Hartley and Professor Ian Reid.

[Register for Event >]https://url.au.m.mimecastprotect.com/s/_sewCE8wmrtg7QlZwfDhyf7aYCr?domain=t.e2ma.net

[https://d31hzlhk6di2h5.cloudfront.net/20260723/2e/8d/c7/1c/1c70baef8b350e384...]

[Web Site]https://url.au.m.mimecastprotect.com/s/eU4VCJyBrGf1J7BjkS7s9fypLgI?domain=t.e2ma.net [LinkedIn]https://url.au.m.mimecastprotect.com/s/_JIHCK1DvKTBojDRQH8twf5iM3k?domain=t.e2ma.net [X]https://url.au.m.mimecastprotect.com/s/alOaCL7EwMfmWJkz0CXuMfy6EFv?domain=t.e2ma.net [YouTube]https://url.au.m.mimecastprotect.com/s/XaPICMwGxOtRrQ2nphoCKf812z9?domain=t.e2ma.net

Copyright © 2026

The University of Sydney

Please add acfr.admin@sydney.edu.aumailto:acfr.admin@sydney.edu.au to your address book or senders safe list to make sure you continue to see our emails in the future.

Cricos Code 00026A TEQSA PRV12057 Disclaimerhttps://url.au.m.mimecastprotect.com/s/_edYCNLJyQUEW2ZgYtwFDfyfNP1?domain=t.e2ma.net | Privacy statementhttps://url.au.m.mimecastprotect.com/s/A02yCOMKzVT086NBDt8HzfGa1So?domain=t.e2ma.net |

_______________________________________________ Everybody mailing list -- everybody@acfr.usyd.edu.au To unsubscribe send an email to everybody-leave@acfr.usyd.edu.au