Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Fully Autonomous Neuromorphic Navigation and Dynamic Obstacle Avoidance
Authors: Xiaochen Shang, Pengwei Luo, Xinning Wang, Jiayue Zhao, Huilin Ge, Bo Dong, Xin Yang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate our system on an autonomous quadrotor using only onboard resources, demonstrating reliable navigation and avoidance of diverse obstacles moving at speeds up to 10 m/s under different light conditions, with energy consumption reduced to 21% compared to traditional architecture. [...] 4 Experiments [...] 4.1 Simulation Experiments [...] 4.2 Real-world Experiments |
| Researcher Affiliation | Collaboration | Xiaochen Shang1, Pengwei Luo1, Xinning Wang1, Jiayue Zhao1, Huilin Ge2, Bo Dong3, Xin Yang1, 1Key Laboratory of Social Computing and Cognitive Intelligence (Dalian University of Technology), Minstry of Education, 2Jiangsu University of Science and Technology, 3Cephia AI |
| Pseudocode | No | The paper describes methodologies in paragraph form and through figures, but it does not contain any explicitly labeled pseudocode or algorithm blocks with structured steps. |
| Open Source Code | No | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [No] Justification: We will release the code upon acceptance. |
| Open Datasets | Yes | Additionally, we introduce the first monocular event-based pose correction dataset with over 50,000 paired and labeled event streams. [...] An open-sourced monocular event-based pose correction dataset with over 50,234 paired and labeled event streams. |
| Dataset Splits | No | For the pose correction dataset, it states "we constructed a novel dataset containing 50,234 event stream pairs... There are four distinct indoor scenarios contained in the dataset". However, no explicit train/test/validation split percentages or sample counts for the neural network training are provided in the main text. Details about "Training details are shown in supplementary note 6." but are not accessible in the main paper. |
| Hardware Specification | Yes | The neuromorphic control framework is implemented on the Speck Neuromorphic SoC [38] and deployed on a small quadrotor... In this work, we use Speck [38], a neuromorphic So C (System on Chip) with 327,000 neurons [39] that could support at most 8 layers of SNNs. |
| Software Dependencies | No | The paper mentions "We use ESIM [51], an event camera simulator in Gazebo, to simulate event camera imaging effects for the quadrotor" but does not specify version numbers for Gazebo or ESIM, nor does it list other software dependencies with versions for their own implementation. |
| Experiment Setup | Yes | We set the closest starting distance of obstacles at 0.2m... we set the snapshot interval at 7.5 meters... We conducted tests at 5-millisecond (ms) intervals and found the best value as 25 ms. Therefore, A1 / A2 should be 1.22, and when τe = 5 ms we get the best result. Under ideal conditions, setting Eth = Ith would enable perfect static event cancellation. However, in practice, sensor noise and firing threshold fluctuations in biological neurons necessitate permitting minor deviations to prevent noise-induced false dynamic responses; here we choose Ith / Eth = 1.2 based on our experimental testing. |