Skip to content
Log in

Author

Ze Ji

h-index 0 View on Semantic Scholar

We have 2 of 5 papers

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

Conference Aug 2026

CardiffNav: A GNSS-centric Multisensory Dataset for Robust Localisation in Diverse and Challenging Environments

Robust state estimation via multi-sensor fusion is strictly constrained in urban environments, driven primarily by Non-Line-of-Sight (NLOS) and multipath interference acting upon Global Navigation Satellite System (GNSS) signals. Current benchmark datasets systematically omit the raw GNSS observables requisite for formulating tightly-coupled mitigation mechanisms. To resolve this, we release CardiffNav, a multi-modal sensor dataset engineered for degraded-environment localisation. The hardware framework synchronises a 128-channel Light Detection and Ranging (LiDAR) sensor, a visual perception array (RGB, stereo, and event cameras), and a 9-axis Inertial Measurement Unit (IMU). Concurrently, the system logs raw multi-constellation, multi-frequency GNSS measurements alongside Intermediate Frequency (IF) signal samples. Recorded trajectories traverse a continuous gradient of signal availability, explicitly documenting the operational transitions across open-sky segments, structural highways, dense urban canyons, and GNSS-denied tunnels. Baseline evaluations indicate that the unconstrained integration of degraded GNSS measurements directly corrupts the coupled state estimate. This dataset consequently provides a rigorous testbed for validating algorithms designed to identify, decouple, and mitigate signal degradation at the measurement level. The complete dataset and benchmark utilities are available at https://github.com/Erika1kuta/CardiffNav.

Shun Liu, Yixin Chen, R. Pullin et al. · 0 citations
Conference Aug 2026

Prompted to Explore: Training-Time-Only LLM Proposals for Sample-Efficient Pushing Grasping Policies

Learning long-horizon robot manipulation remains difficult and time-consuming, especially under sparse rewards due to inefficient exploration and reward assignment. We present a minimal integration of large language models (LLMs) with reinforcement learning (RL) in which the LLM is used strictly as an online action proposer during early training to help with the RL agent. Given a task description and a compact scene abstraction, the LLM outputs a single parameterised primitive (pushing or grasping) that is validated by lightweight safety checks and mixed with policy actions via an epsilon-greedy scheduler. Besides, we study two simple policies that preserve the same low-level learner (grasping-biased model): (i) pushing-first reward shaping that temporarily increases the relative reward of pushing early in an episode, and (ii) a Proximal Policy Optimization (PPO) high-level switch that explicitly selects between push and grasp while Soft Actor-Critic (SAC) handles low-level parameterisation. In simulation, LLM-guided exploration accelerates early learning, and the proposed methods help further mitigate grasping overuse in the cluttered environment. Besides, the LLM-guided RL agent could achieve a higher success rate at 89.5% compared with the baseline method at 68%.

Meiyuan Gong, Yan Gao, Ze Ji · 0 citations