Johns Hopkins University SPIKE Lab Electrical & Computer Engineering

Research Areas

Event-based, continuous-time perception for systems that see and act at the speed of the physical world.

Event-based Vision & 3D Computer Vision

Our lab develops algorithms for event-based cameras, which represent a paradigm shift in computer vision, and uses them to recover the 3D structure of the world. We focus on continuous video reconstruction, high-speed vision, and event-based 3D reconstruction. For example, EvAC3D converts event-based apparent contours into accurate 3D models, enabling real-time reconstruction of dynamic scenes.

  • Continuous color video reconstruction
  • Event-based 3D reconstruction
  • Motion field estimation
  • Continuous visual hull computation
  • Low-latency visual processing
  • Real-time 3D modeling
EvAC3D: 3D models from event-based apparent contours · ECCV 2022 Oral

Neuromorphic AI & Machine Learning

Neuromorphic sensors and brain-inspired computation promise perception that is fast, sparse, and power-efficient. We build the models, data, and tools needed to learn from neuromorphic signals at scale. For example, EVIS is a physics-grounded event camera plugin for NVIDIA Isaac Sim that simulates high-rate events, with HDR support, sensor noise models, and motion blur, to generate training data for downstream tasks.

  • Neural event processing
  • Physics-grounded event simulation
  • Synthetic data for event-based learning
  • Efficient, low-latency inference
EVIS: simulated robot views and the events they produce, in NVIDIA Isaac Sim

Active Perception

Rather than passively recording whatever passes in front of the sensor, our robots decide where to look. We plan sensing trajectories that gather the most informative views while respecting what the robot can physically do. For example, TRACE is the first trajectory-level ergodic formulation for active Gaussian scene reconstruction: it gains about 1.5 dB PSNR over next-best-view baselines and runs directly on real robots, including the Unitree GO2 and Franka FR3.

  • Ergodic trajectory optimization
  • Active 3D scene reconstruction
  • Information-driven exploration
  • Dynamics-aware view planning
TRACE: ergodic camera trajectories (orange) for active scene reconstruction

Robotics

We bring event-based perception onto real robots. Low-latency event cameras let robots react to fast-moving objects, and our work spans high-speed object catching, low-latency perception for control, and large multi-robot, multi-sensor datasets.

  • High-speed object catching
  • Low-latency perception for control
  • Multi-robot, multi-sensor datasets
  • Real-world deployment
EV-Catcher: high-speed object catching with low-latency event-based neural networks (1/8× speed)

Current Projects

Lab code on GitHub →

arXiv preprint · 2026

TRACE

Ergodic trajectory optimization for active scene reconstruction: the first trajectory-level ergodic formulation for active Gaussian scene reconstruction. Footprint-aware ergodic search gains +1.5 dB PSNR over baselines and runs directly on real robots, including the Unitree GO2 and Franka FR3.

Open-source software · 2026

EVIS

A physics-grounded event camera plugin for NVIDIA Isaac Sim. High-rate event simulation accelerated by motion-vector frame interpolation, with HDR support, sensor noise models, motion blur, and tools for downstream training.

ECCV · 2026

Match-Any-Events

Zero-shot, motion-robust feature matching across wide baselines for event cameras: the first generalizable matching model with events.