New Presenting at NeurIPS 2025

Korneel Van den Berghe

ML Robotics Engineer & Researcher

Building intelligent agents that interact with the physical world. 4+ years of experience in reinforcement learning, autonomous systems, and neuromorphic computing.

Currently AI Engineer at Harbor AI in NYC, developing models for dynamic insurance optimization. Previously visiting researcher at Harvard's Edge Computing Lab.

Research & Development

Projects

Exploring the intersection of neuromorphic computing, reinforcement learning, and autonomous systems.

Crazyflie drone flight with SNN controller

NeurIPS 2025 Oral

Adaptive Surrogate Gradients for SNNs

Training spiking neural networks for real-world drone control. Covers gradient scheduling, the warm-up period challenge, and zero-shot sim-to-real transfer.

Part 1: Gradients Part 2: RL & Deployment

ICNCE 2024

SNNs for Reinforcement Learning Control

Foundational work comparing spiking vs non-spiking networks in A2C, from CartPole benchmarks to real-world drone landing on the Bebop 2.

NeuroBench: Closed-Loop Benchmarking

My contribution to the NeuroBench framework: metrics and evaluation protocols for neuromorphic control tasks. Published in Nature Communications.

A2Perf: Autonomous Agents Benchmark

Benchmarking suite for autonomous agents with real-world domains: chip floorplanning, web navigation, and quadruped locomotion.

Curriculum Vitae

Experience in ML research, robotics engineering, and aerospace systems.

Experience

Lead ML Researcher & Engineer

Aug 2024 – Present

Harbor AI

New York City, USA

Leading RL models for dynamic insurance price optimization
Interpretability research for explainable risk analysis
Building MLOps infrastructure for $100M+ valuation scale

Visiting Researcher, ML

Aug 2023 – Jul 2024

Edge Computing Lab, Harvard University

Boston, USA

Co-authored and maintain NeuroBench benchmarking framework
Contributed to A2Perf RL benchmarking suite
RL-based pruning for TinyML drone controllers

ML Researcher

Aug 2020 – Dec 2024

Honours Programme MSc, TU Delft

Delft, NL

Neuromorphic MAV controllers using event data and optic flow
Deep Q-learning for low-power RL on edge devices
Topology optimization for compliant wing morphologies

Structural & Control Systems Engineer

Sep 2022 – Dec 2023

AeroDelft (Hydrogen Aircraft)

Delft, NL

Control systems in C++ for first student hydrogen aircraft
CAD to flight-ready hardware prototyping

Education

M.Sc. Aerospace Engineering (Honours)

2022 – 2024

TU Delft

Specialization: Control and Simulations

B.Sc. Aerospace Engineering (Honours)

2019 – 2022

TU Delft

Publications

Adaptive Surrogate Gradients for Sequential RL in Spiking Neural Networks

K. Van den Berghe, S. Stroobants, V. J. Reddi, G.C.H.E. de Croon

Oral Presentation at NeurIPS 2025

NeuroBench: Advancing Neuromorphic Computing through Collaborative Benchmarking

J. Yik, K. Van den Berghe, C. Frenkel, V. J. Reddi, et al.

Nature Communications, NICE 2024

NeuroBench: Closed Loop Benchmarking

K. Van den Berghe, J. Yik, C. Frenkel, V. J. Reddi, et al.

Neuromorphics Netherlands 2024, ASPLOS 2025

A2Perf: Benchmarking Autonomous Agents in Real-World Domains

I. Uchendu, J. Jabbour, K. Van den Berghe, et al.

Under review, AAMAS 2026

Control with SNNs Trained with RL Using Surrogate Gradients

K. Van den Berghe, S. Stroobants, G.C.H.E. de Croon

ICNCE 2024

Skills

Reinforcement Learning Neuromorphic Computing Machine Learning Autonomous Systems TinyML Edge Computing Python C++ PyTorch MLOps Control Systems Robotics

From the Archives

Fun Projects

A collection of fun projects from my past that don't fit my current research path, but I enjoyed doing.

2012

The Chainsaw Bike

At age 11, I attempted to build a motorized bike using a broken chainsaw from my shed. Dreams of a motorcross bike, reality of a mountain bike with an engine strapped to it.

Read the Story