Research direction and evidence

Research direction · Spatial intelligence

How can intelligent systems understand a world they only partly observe?

My long-term direction is spatial intelligence: helping intelligent systems understand the physical world—its properties, relationships and changes—well enough to reason and act. I approach this through observation. Sensors provide partial, noisy and different evidence; my published work studies how to recover useful physical quantities from it. Maintaining that understanding over time and using it in decisions are the next questions.

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.
RGB plant observation from Proteus, with the paper's original red annotationsCamera
mmWave SAR observation of the plant from ProteusmmWave
One plant. Two ways of seeing.

RGB and mmWave SAR examples from Proteus. Each reveals a different part of the same physical scene. Proteus

Toward spatial intelligence

Research roadmap

Published work establishes sensing foundations; current work moves from representation to understanding and action for one observer. The roadmap then extends the loop to many observers sharing one space—open research questions, not results.

  1. 2022–2023

    Published foundations

    Recovering a signal

    Biomimetic sonar and mmLeaf introduced me to indirect observation: finding structure in echoes and measuring leaves with radio waves.

    Early work
  2. 2024

    Published foundations

    Learning from complementary views

    Hydra combined radar and camera observations. That made the representation—not simply the choice of sensor—the next question.

    Hydra
  3. 2025

    Published foundations

    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
  4. Current research

    In progress · public preprint

    From a snapshot to a continuing record

    Driving assessment motivates a shift from isolated measurements to behavior understood over time and in context. The public preprint sets out design principles and research opportunities; a completed monitoring evaluation remains to be established.

    Driving assessment
  5. 2026

    Public preprints · one manuscript

    From representation to understanding and action

    SPIRIT reconstructs explicit 3D surfaces from sparse radar measurements. Mímir lets an LLM agent propose irrigation decisions while physics keeps authority; Snotra revises its world model only when evidence supports the change.

    Preprints & manuscript
  6. Next

    Future agenda

    Shared spatial intelligence

    Let several observers combine what they see, believe and do: when views complement each other, how shared beliefs are revised, and who may act. AURA’s multi-view driving setting is the first test.

    Research vision

Selected research

Three published sensing systems and a driving-assessment preprint. Each asks a specific question and provides a different kind of evidence.

Hydra radar and camera testbed deployed in a crop field

Published research / ACM MobiCom 2024

Hydra

Measuring leaf wetness with radar and vision.

Combining radar and camera observations for leaf wetness sensing in changing field conditions.

Read the study
Proteus radar sensing prototype and plant target

Published research / ACM SenSys 2025

Proteus

Learning with vision, sensing with radar.

Transferring visual knowledge to mmWave representations, with more informative radar imaging.

Read the study
Adonis mmWave radar experiment with a plant and scan geometry

Published research / IEEE INFOCOM 2025

Adonis

Measuring how wet a leaf is.

Fine-grained leaf wetness estimation from complementary radar signal views and calibration.

Read the study
Driving simulator with steering wheel, pedals and a road scene on the monitor

Current research / arXiv preprint 2026

AURA

Driving assessment over time and in context.

Exploring continuous, contextualized driving assessment for older adults.

Read the study

Future agenda

From one observer to shared spatial intelligence

Today’s intelligence lives on islands: a field sensor sees one leaf, a car one viewpoint, a robot one room, a language model words rather than places. One observer is a sparse aperture on the world. My next step is to let observers combine what they see, believe and do—without giving up local authority.

Observe together

When do several partial views determine what one cannot?

Share evidence across sensors and agents, and measure what becomes identifiable only when views are combined.

Challenge: alignment, timing and bandwidth across heterogeneous sensors.

Believe together

How should shared beliefs be revised—and when should they not be?

Maintain a common model of the space with explicit uncertainty, admitting revisions only when evidence supports them.

Challenge: avoiding double-counted evidence and conflicting sources.

Act together

Who may act on shared understanding?

Coordinate decisions across agents while each keeps physical authority bounded and local.

Challenge: safety, accountability and privacy when consequences are physical.