BioMorphic Intelligence Lab

Biologically inspired solutions for aerial robotics

Aerial robots are now ubiquitous. Thanks to their nimbleness, manoeuvrability, and affordability, drones are used in many sectors to monitor, map, and inspect. As a next step, flying robots offer more when interacting with their surroundings via anthropomorphic-like manipulation capabilities. Some overarching challenges remain for this new class of aerial robots, and solutions inspired by biology can be implemented across three key areas for robot performance: sensing their environment, processing this information, and acting upon the results.

SENSE

Bio-inspired perception (e.g., visual or tactile feedback) can provide the drone with information on its environment, mimicking animalsā€™ sensory feedback. Using retina-like event cameras, drones can avoid obstacles and detect objects at a fraction of the power and latency of conventional hardware and algorithms. Enhancing tactile feedback can also prompt different behaviors in response to different force stimuli.

THINK

Bio-inspired, brain-like models from Neuromorphic AI can help lower the computational load and speed up sensory data processing for navigation. This boosts real-time control and autonomy. Compliance embedded in the control of the robot also favors safe and robust interaction with unknown environments and targets.

ACT

Bio-inspired design and materials make the drone's body fit for interaction with unknown objects and enable a safe response to external disturbances. Robot morphology can be inspired by flying animalsā€™ shape, configuration, and materials. Together, these features create embodied intelligence and can partially offset the behavior complexity handled by the brain.

The BioMorphic Intelligence Lab aims to tackle robustness and efficiency challenges for interacting drones, using biologically inspired solutions for both the 'body' and the 'brain' and applying embodied intelligence and neuromorphic AI techniques.

The BioMorphic Intelligence Lab is part of the TU Delft AI Labs programme.

Collection of training data is an integral part of modern AI research. A helmet mounted with two retina-inspired event cameras can be used to mimic the egocentric, stereo vision of mammals, and the recorded videos can be used for biologically-inspired machine learning.

If we gave drones human-like fingers, they would be able to help humans in difficult situations at height. By embodying the sense of touch, drones can feel the environment in the same tactile way as humans, in the absence of light.

In nature, animals use their limbs for grasping and walking. By imitating biological functions, drones with lightweight limbs can better cope with impacts but also use their limbs for locomotion. This way,  drones will become flexible to operate in different environments.

The Team

Directors

PhD students

Associated faculty

Education

Courses

  • Every Semester: Design Synthesis Exercise (DSE) for Bachelor Students.  The projects are always within the field of ā€œDrones for Environmental Monitoringā€.
  • Project supervision in the Deep Learning (CS4240) course for MSc Computer Science.
  • Project supervision in the Research Project (CSE3000) course for BSc Computer Science.
  • Guest lecture on "Event-Based Vision" in the Computer Vision by Deep Learning (CS4245) course for MSc Computer Science
  • Guest lecture on "Neuromorphic Computing" in the Machine Learning 2 (CS4230) course for MSc Computer Science.
  • Supervision of MSc Graduation Projects for the Computer Science program (IN5000), see below for open project topics.
  • Guest Lecture on ''Physical Intelligence for Interaction'' (AE4350) course for MSc in Aerospace Engineering
  • MSc Course: Physical Interaction for Aerial and Space Robots (AE4324)

AI minor and AI Master Block programs

  • Project supervision for the Capstone Applied AI Project (TI3150TU)
  • Fundamentals of AI (IFEEMCS520100) and
  • Interdisciplinary Advanced AI Project (IFEEMCS520200) courses.

Master projects

Ongoing

  • Soft Arerial Manipulation
  • Tactile Aerial Navigation
  • Dynamic Perching
  • Compliant Landing

Finished

For a more detailed list of Thesis Topics, please see the following link (which is only accessible if you have a TU Delft account)

Partners