Fengyi Wang

Fengyi Wang

Auto-generated profile, based on publicly available information

Fengyi Wang

Institute for Cognitive Systems Technical University of Munich Munich Germany

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 June 2021 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
FWFengyi Wang
Fengyi Wang
JGJulio Guadarrama-Olvera
Julio Guadarrama-Olvera
GCGordon Cheng
Gordon Cheng
In this letter, we present a fast method for autonomously planing manipulation tasks for mobile manipulators. The planner defines an optimal order to perform pick-and-place operations for taking objects from a cluttered scene to specific deposit areas considering both, manipulator and mobile base motion. Our method first examines the grasping feasibility of the objects with an inverse reachability map. Then, it defines all the placing locations for the objects and analyses the corresponding preconditions to reach them. Finally, it defines a sequence for rearranging the objects that minimize the execution time. We take advantage of the environment's underlying combinatorial structure to define the shortest path. In this work, we consider both the monotone case, where each object may be moved at most once, and the non-monotone cases. An experimental evaluation on the Human Support Robot (HSR) shows the effectiveness of our solutions and the scalability of our method as the number of objects increases for both cases. Tests on monotone problem instances with 20 objects show the proposed method can save up to 17% base traveling time when comparing to baseline methods at the cost of less than 9 seconds planning time. In a test on a simple non-monotone instance, the proposed method further reduces the total execution time by 25% by minimizing the total number of actions in a few seconds of planning time. The characteristic speed in re-planning makes our method suitable for online usage.

20 April 2023 in IEEE Robotics and Automation Letters
Verified
Journal Article
Open access
FWFengyi Wang
Fengyi Wang
JGJulio Guadarrama-Olvera
Julio Guadarrama-Olvera
NTNitish Thakor
Nitish Thakor
GCGordon Cheng
Gordon Cheng
In this letter, we present a method for realizing the heat-evoked nociceptive withdrawal reflex (NWR) in the upper limb of a humanoid robot so that it can avoid the potential damage caused by noxious heat. We use a spiking neural network whose structure, encoding scheme, and form of information transmission mimic the reflex arc in humans to improve bio-plausibility. The proper synaptic strengths between the sensory neurons and the interneurons in the first two layers are learned using the bio-plausible reward-modulated spike timing-dependent plasticity learning algorithm. By monitoring the spikes from the motor neuron in the third layer, a reflex matching the intensity of the stimulation can be evoked. Experimental evaluations show that noxious heat stimulation can be detected online and evoke the NWR. The experiments on a full-size humanoid robot show that the method enables robots to avoid potential damage robustly with proper NWR, depending on the site and intensity of the stimulation. We also verify that the method takes advantage of the intrinsic characteristics of its neuromorphic encoding scheme to reproduce essential features of the NWR, e.g., spatial summation effect and temporal summation effect in humans. The improved bio-plausibility and the capability to reproduce the human-like feature.

16 September 2026 in Advanced Robotics Research
Verified
Journal Article
Open access
FWFengyi Wang
Fengyi Wang
NTNitish Thakor
Nitish Thakor
GCGordon Cheng
Gordon Cheng
Service robots often struggle to retrieve objects from occluded or poorly illuminated spaces without reliable camera observations. This work presents a bio‐inspired approach combining proximity‐guided adaptive thigmotactic exploration with neuromorphic recognition of multimodal contact signals. Distributed active‐infrared proximity sensors around the end effector provide sparse, low‐spatial‐acuity optical measurements that regulate stochastic motion and boundary following during collision‐aware exploration. After a candidate object is detected, receptor‐informed encoders transform force, temperature, and wrist force–torque sequences acquired during grasping and lifting into spike trains for classification by a spiking neural network (SNN). Across 50 shared initial–target configurations with five repetitions each, adaptive thigmotaxis reduces mean traveled distance by 31.65% (95% CI: 20.38%–41.82%) relative to Lévy walk. The SNN achieves mean test‐set accuracies of 97.46% and 100% on two collected datasets, and object retrieval succeeds in 49 of 50 robot trials. With input preparation excluded for both GPU classifiers, SNN inference reduces total and dynamic energy by 60.1% and 70.3%, respectively, relative to a long short‐term memory network. Under the no‐I/O core‐load condition on Loihi 2, the quantized SNN requires 0.20 mJ per sequence.