Hydra
Two imperfect views. One better measurement.
Combining radar and camera observations for leaf wetness sensing in changing field conditions.
Read the study
Ph.D. candidate · Computer Science & Engineering
Michigan State University
Yimeng Liu
Understanding a world we can only partly see.
I study how intelligent systems turn incomplete physical observations into reliable understanding—starting with sensing, and looking toward reasoning and action.
Camera
mmWaveRGB and mmWave SAR examples from Proteus. Each reveals a different part of the same physical scene. Proteus
A measurement is a view. It is never the whole world.
A camera sees appearance. Radar responds to physical structure and surface water. Light changes, leaves move, and observations disappear. A useful system must reason about both the evidence it has and what that evidence leaves out.
- Partial
- A sensor observes only what its viewpoint and physics allow.
- Different
- Two modalities can observe the same thing without measuring the same property.
- Changing
- An estimate must be reconsidered as the world and its observations change.
The question grew with the work.
My background is in mathematics. Sensing brought me to the gap between an observation and the physical state we want to know.
-
Recovering a signal
Biomimetic sonar and mmLeaf introduced me to indirect observation: finding structure in echoes and measuring leaves with radio waves.
Early work -
Learning from complementary views
Hydra combined radar and camera observations. That made the representation—not simply the choice of sensor—the next question.
Hydra -
Making the representation useful
Proteus studies what a radar can learn from a camera. Adonis asks how to measure degrees of wetness, with calibration for field conditions.
Proteus & Adonis -
From a snapshot to a continuing record
Driving assessment brings time and context into the problem. My current questions concern how understanding persists as observations and conditions change.
Driving assessment
Ideas, built and tested.
Three published sensing systems, followed by a current application that asks a different kind of question.
Proteus
Learn across modalities. Sense with radar.
Transferring visual knowledge to mmWave representations, with more informative radar imaging.
Read the study
Adonis
Beyond wet or dry. Measure how much.
Fine-grained leaf wetness estimation from complementary radar signal views and calibration.
Read the study
Driving assessment
A continuing record. A human context.
Exploring continuous, contextualized driving assessment for older adults.
Read the studyBeyond a good snapshot.
The independent research program I want to build connects physical measurement, persistent understanding, and responsible decisions. These are future research directions, not completed capabilities.
Understanding with incomplete evidence
What physical state can we infer—and what must remain unknown?
Develop representations that connect heterogeneous observations to physical quantities and make uncertainty explicit.
Understanding that lasts
How should a system revise its beliefs when observations stop or the environment changes?
Study continuity, change and uncertainty over time, with evaluation that exposes missing observations and shifting conditions.
Understanding worth acting on
When is an estimate reliable enough to support a decision?
Connect model reliability to the consequence of an error, including when a system should ask for more evidence or defer to a person.
How I approach a problem.
- Begin with the quantity.
- Define what we need to know about the world before choosing a model.
- Keep the observation in view.
- Ask which physical effects produced the measurement, and which explanations it can distinguish.
- Test where the assumption breaks.
- A useful evaluation includes the conditions that make a method stop working.
What changes between a measurement and understanding?
A research perspective on observation, representation and the questions still open.
Read the perspectiveThe academic record.
Recent papers and public research resources. The complete academic record is one level deeper.
-
2026arXiv'26
AIoT-Based Continuous, Contextualized, and Explainable Driving Assessment for Older Adults
-
2026IMWUT'26
-
2026CEA'26
Research resources
Hydra-Bench — multimodal leaf-wetness data & benchmark preprint
From mathematics to the physical world.
I am a Ph.D. candidate in Computer Science and Engineering at Michigan State University, advised by Zhichao Cao. Before MSU, I studied mathematics at Virginia Tech and worked on biomimetic sonar with Rolf Müller.
Ph.D. candidate, Computer Science and EngineeringMichigan State University
B.S., MathematicsVirginia Tech
I welcome conversations about multimodal sensing, physical representations, and reliable intelligent systems—and collaborations that connect these questions to real settings.
liuyime2@msu.edu