People

M2020 Surface Simulator (SSim) (2017-present)

• Developed flight software (FSW) in-the-loop simulation for sequence validation for rover planner tools
  

• Research underwater perception with sonar sensors for creation of world model occupancy grids
  
• Developed ROS-based, closed loop simulation with models of sonar sensors for world model, localization, and planner development and testing
  

• Led software and hardware design using low SWaP co-processors for short range (< 2km) hazard avoidance that uses 4 nVIDIA Tegras for image acquisition, stereo processing, and contact tracking
  
• Implemented CUDA version of stereo correlator that boosts performance 3x compared to ARM-based implementation
  

• Led software development of stereo algorithms and hardware infrastructure for using high resolution (3904x3904) cameras for long range, radar-free intelligence, surveillance, and reconnaissance (ISR)
  
• Performed profiling to pinpoint bottlenecks and optimized stereo pipeline using multi-cores, multi-threads, and vectorized instructions to boost performance to 1.7Hz processing from 0.5Hz with dual 12-core Xeon
  

• Modified and tuned daytime perception pipeline of low-cost perception head to integrate IR with EO and Lidar to mitigate range data dropouts in the EO data due to textureless areas in over-and under-exposed areas
  
• Developed 2D ground truth tool that quantitatively showed that with IR, fewer false positives and negatives are seen in the perception system
  

• Developed Planetary Illumination Modeling (PLUM) to study lighting conditions on ocean worlds & icy planets for implications on future robotic perception systems
  
• Modeling indirect and direct lighting for a general planetary body and reflectance bodies, PLUM leverages the JPL NAIF SPICE library to provide high fidelity solar system geometry for the basis of our lighting models
  
• Output data products include total, direct, and indirect irradiance plots for a given latitude and longitude on the planetary surface for a given time period and time step
  

• Led development of passive, night-time stereo algorithms using IR cameras to overcome low SNR that yields poor disparity density
  
• Developed multi-resolution consistency check method to reduce false positives from mixed pixels and repeated textures; verified through extensive field trails
  
• Leveraged AVX2 vectorization to boost stereo correlator performance by 35% to maintain 10Hz processing rate in ROS-based framework
  

• Evaluated RabbitMQ and Kafka message brokering frameworks on how well they can enable a real-time, scalable, and robust telemetry framework for multiple consumers across heterogeneous platforms
  
• Integrated message brokering publish-subscribe service with the Sampling Lab Universal Robotic Manipulator (SLURM) server code to publish 130 channels of telemetry used during environmental testing