Robotics and AI researcher currently at the Chinese Academy of Sciences (CASIA), working on Embodied Intelligence — robot learning, safety benchmarks for Vision-Language-Action (VLA) models, and continual learning with Spiking Neural Networks.
- 🎓 UCL — MSc Robotics and AI, Distinction (76.71/100)
- 🎓 UESTC — BEng Robot Engineering, GPA 3.93/4.0, National Scholarship
- 🧠 Interests — Robot perception / manipulation / interaction, RL, biologically-inspired AI
| Category | Tools & Technologies |
|---|---|
| Languages & Frameworks |     |
| Simulators | Isaac Lab · MuJoCo · DMControl · RoboTwin |
| Algorithms | RL (PPO, DreamerV3) · Imitation Learning (DAgger) · NMPC · Dynamic Graph SNN |
Unified diagnostic safety benchmark for Vision-Language-Action models.
- Taxonomy of 13 categories covering physical, linguistic, and perceptual safety.
- 66 safety-enhanced scenarios in RoboTwin for structural and visual perturbations.
- Continual learning framework based on Dynamic Spiking Neural Networks for task-agnostic structural adaptation.
- SNN-RL training stack with LIF neurons and surrogate gradients on DMControl.
- Interdisciplinary work aligning AI self-recognition with macaque/human bodily-self experiments.
- Isaac Lab + DreamerV3 training for multi-perspective self-recognition.
- Adaptive architecture using local structural plasticity to dynamically grow/prune nodes during training.
- 📄 Published — arXiv:2512.12713.
- MSc with Distinction, UCL — Computer Vision 89.03, Legged Robotics 80.47
- National Scholarship (2021) & Outstanding Engineer award (2023)
- ✉️ Email — yiyang.jia.24@ucl.ac.uk
- 📄 arXiv — 2512.12713
- 🌐 English — IELTS 7.5
"Exploring the intersection of biological plasticity and robotic autonomy."