Institution profile

Centre for Perceptual and Interactive Intelligence

Academic institutionasia · hk
Official website
Research library18linked papers
Opportunities0open roles
Selected work

Representative Papers

HPSD: Hybrid-Policy Self-Distillation for Text-Image-to-Video Diffusion Models

Aug 13, 2026

This work aims to effectively transfer the superior generation capabilities of Text-Image-to-Video (TI2V) models—exhibited when conditioned on high-quality initial frames or detailed textual prompts—to their underlying Text-to-Video (T2V) task. To this end, the authors propose a hybrid self-distillation framework in which a single model acts as both teacher (in TI2V mode) and student (in T2V mode). The approach leverages off-policy trajectory anchoring, local policy optimization, and velocity-level supervision signals to enable precise policy correction while avoiding condition-state mismatches. Notably, this method is the first to integrate privileged priors with online policy fine-tuning, significantly enhancing base T2V generation quality without requiring additional data, while also further improving TI2V performance—thereby comprehensively strengthening the model’s video synthesis capabilities.

0 citationsRead paper

Trajectory Optimization for Collision-Aware Redundant Robotic Multi-Axis Additive Manufacturing by Constrained Gradient Projection

Jun 29, 2026

This work addresses the challenge of trajectory optimization for long-range paths in redundant robotic multi-axis additive manufacturing under stringent deposition position constraints and time-varying collision constraints. The authors propose a collision-aware trajectory optimization framework that models the kinematic relationship between the nozzle and workpiece using relative Jacobians, captures dynamic geometric evolution through differentiable signed distance fields (SDFs), and enforces hard deposition position constraints via iterative projection onto the self-motion manifold. Optimization efficiency is enhanced by restricting gradient updates to the tangent space. Experimental results on an 8-degree-of-freedom platform demonstrate sub-10-micron average nozzle positioning error, a 77.6% reduction in peak joint jerk, complete avoidance of collisions and posture violations, up to a 10.2× speedup over an SQP baseline, and successful fabrication of complex unsupported structures.

0 citationsRead paper
Recent publications

Latest Papers

HPSD: Hybrid-Policy Self-Distillation for Text-Image-to-Video Diffusion Models

Aug 13, 2026

This work aims to effectively transfer the superior generation capabilities of Text-Image-to-Video (TI2V) models—exhibited when conditioned on high-quality initial frames or detailed textual prompts—to their underlying Text-to-Video (T2V) task. To this end, the authors propose a hybrid self-distillation framework in which a single model acts as both teacher (in TI2V mode) and student (in T2V mode). The approach leverages off-policy trajectory anchoring, local policy optimization, and velocity-level supervision signals to enable precise policy correction while avoiding condition-state mismatches. Notably, this method is the first to integrate privileged priors with online policy fine-tuning, significantly enhancing base T2V generation quality without requiring additional data, while also further improving TI2V performance—thereby comprehensively strengthening the model’s video synthesis capabilities.

0 citationsRead paper

Trajectory Optimization for Collision-Aware Redundant Robotic Multi-Axis Additive Manufacturing by Constrained Gradient Projection

Jun 29, 2026

This work addresses the challenge of trajectory optimization for long-range paths in redundant robotic multi-axis additive manufacturing under stringent deposition position constraints and time-varying collision constraints. The authors propose a collision-aware trajectory optimization framework that models the kinematic relationship between the nozzle and workpiece using relative Jacobians, captures dynamic geometric evolution through differentiable signed distance fields (SDFs), and enforces hard deposition position constraints via iterative projection onto the self-motion manifold. Optimization efficiency is enhanced by restricting gradient updates to the tangent space. Experimental results on an 8-degree-of-freedom platform demonstrate sub-10-micron average nozzle positioning error, a 77.6% reduction in peak joint jerk, complete avoidance of collisions and posture violations, up to a 10.2× speedup over an SQP baseline, and successful fabrication of complex unsupported structures.

0 citationsRead paper