arXiv subjects

Yunhao Zhao

Publications and source records attributed to Yunhao Zhao.

7 recordsLinked to original sources

arXiv2026

PL-NBA: A Possession-level Universal Basketball Video Dataset Supporting Multiple Visual Understanding Tasks

Visual understanding in sports has emerged as a hot topic in computer vision in recent years. Most existing basketball video datasets adopt single action or activity as sample, which can neither preserve the temporal continuity of game events nor support complex tasks such as action anticipation. To address this issue, this paper constructs the first possession-level basketball video dataset (PL-NBA), in which each sample is composed of a complete NBA offensive possession. Collected from 60 NBA games, PL-NBA contains 11,000 valid offensive possession clips and 31,567 annotated events with player names, captions, event types and timestamps. Each video clip includes multiple events and preserves the continuity of events, which is helpful for analysis of tactic. Experiment is conducted on multiple visual understanding tasks, including event recognition, video captioning, temporal action localization and action anticipation. Experimental results show that existing methods achieve limited performance on above four tasks, demonstrating that PL-NBA is a challenging benchmark for sports video understanding.

cs.CV↗

arXiv2026

Skills in Weights, Memory in Code: Hybrid Learning for Memory-Dependent Robot Manipulation

Modern vision-language-action (VLA) policies have acquired broad manipulation skills, but typically generate each action chunk from the current observation or a short fixed-length history. However, real-world manipulation is often non-Markovian, requiring robots to retain and reason over task-relevant information from long-horizon interaction histories to determine the next action. To address this challenge, we propose HyMeS, a hybrid learning framework that leverages the reasoning and memory-management capabilities of coding agents to steer a Markovian VLA for memory-dependent manipulation. Specifically, HyMeS learns low-level motor skills through gradient-based imitation learning, while a coding agent acquires high-level memory-management strategies through heuristic learning by iteratively updating an executable heuristic system from rollout feedback. Furthermore, we close the loop between steering and execution through multimodal stage-completion verification, which updates memory using proprioceptive signals and multi-frame VLM judgments. Compared with end-to-end memory-augmented VLAs, HyMeS requires demonstrations only for reusable motor skills rather than for every history-dependent task configuration, enabling data-efficient compositional generalization. On RoboMemArena, HyMeS improves mean cumulative success from 52.5% to 66.2% and mean task success from 41.3% to 60.1% over pi0.5, while outperforming PrediMem by 4.5 points in cumulative success and 14.5 points in task success.

cs.RO↗

arXiv2022

Phase Transformations During Continuous Cooling in Inconel 718 Alloys Manufactured by Laser Powder Bed Fusion and Suction Casting

Understanding alloy phase transformations during continuous cooling is important for post-processing design and optimization. In this work, continuous-cooling-transformation (CCT) diagrams of Inconel 718 alloys manufactured by laser powder bed fusion (LPBF) and suction casting are developed under different homogenization conditions. Unlike the available CCT diagrams in the reported studies, no gamma double prime and gamma prime precipitates can be observed. NbC and delta are determined to be the precipitates after cooling from the gamma matrix. Importantly, homogenization time and manufacturing methods are found to affect the Nb homogeneity in the matrix near NbC particles and thus significantly influence the precipitation process of the delta phase, which has a high content in Nb. In the alloys with high Nb homogeneity, the nucleation process mainly contributes to the precipitation, whereas in the alloys with low Nb homogeneity, the precipitation is primarily associated with the growth process. Subgrains are found to form after cooling at 0.1 K/s and can cause the highest hardness in samples. This work provides a new viewpoint on the study of processing-structure-property relationships during cooling in Inconel 718 and is beneficial to the development of alloy post-processing strategies.

cond-mat.mtrl-sci↗

arXiv2020

A Comparative Analysis of Inconel 718 Made by Additive Manufacturing and Suction Casting: Microstructure Evolution in Homogenization

Homogenization is one of the critical stages in the post-heat treatment of additive manufacturing (AM) component to achieve uniform microstructure. During homogenization, grain coarsening could be an issue to reserve strength, which requires careful design of both time and temperature. Therefore, a proper design of homogenization becomes particularly important for AM design, for which work hardening is usually no longer an option. In this work, we discovered an intriguing phenomenon during homogenization of suction-cast and AM Inconel 718 superalloys. Through both short and long-term isothermal heat treatments at 1180°C, we observed an abnormal grain growth in the suction-cast alloy but continuous recrystallization in the alloy made by laser powder bed fusion (LPBF). The grain size of AM samples keeps as small as 130 μm and is even slightly reduced after homogenization for 12 hours. The homogeneity of Nb in the AM alloys is identified as the critical factor for NbC formation, which further influences the recrystallization kinetics at 1180°C. Multi-type dislocation behaviors are studied to elucidate the grain refinement observed in homogenized alloys after LPBF. This work provides a new pathway on microstructure engineering of AM alloys for improved mechanical performance superior to traditionally manufactured ones.

cond-mat.mtrl-sci↗

arXiv2020

A new high-throughput method using additive manufacturing for alloy design and heat treatment optimization

Many alloys made by Additive Manufacturing (AM) require careful design of post-heat treatment as an indispensable step of microstructure engineering to further enhance the performance. We developed a high-throughput approach by fabricating a long-bar sample heat-treated under a monitored gradient temperature zone for phase transformation study to accelerate the post-heat treatment development of AM alloys. This approach has been proven efficient in determining the aging temperature with peak hardness. We observed that the precipitation strengthening is predominant for the studied superalloy by laser powder bed fusion, and the grain size variation is insensitive on temperature between 605 and 825 Celcius. This new approach can be applied to post-heat treatment optimization of other materials made by AM, and further assist new alloy development.

cond-mat.mtrl-sci↗

arXiv2020

Cyclic Re-austenitization of Copper-bearing High-Strength Low-Alloy Steels Fabricated by Laser Powder Bed Fusion

For the first time, cyclic re-austenitization is carried out for additively manufactured high-strength low-alloy (HSLA) steels in order to refine the microstructure by reducing the prior austenite grain (PAG) size. In this work, HSLA-100 steels processed using laser powder bed fusion (LPBF) technique are subjected to several cycles of re-austenitization using quenching dilatometry. Microstructure characterization for every cycle revealed the presence of bainite, martensite and martensite/austenite (M/A) islands. From the analysis of the dilatometry curves and extensive microstructure characterization, it was found that till the 2nd cycle of re-austenitization, both PAG size and martensite start (Ms) temperature get reduced, while the amount of bainite transformed decreased and the retained austenite content increased. Concomitantly, the highest microhardness along with peak nanohardness of the constituent phases was achieved at the 2nd cycle. Conversely, from the 3rd cycle, the microhardness, as well as the nanohardness of the constituent phases, are found to decrease due to an increase in the PAG size. This behavior is in contrast to the general tendency where a saturation limit is reached after the peak refinement is achieved. It is found that retained austenite can act as a pinning particle to obstruct the PAG boundary movement and its fraction is found to decrease from the 3rd cycle. Hence, the increase in PAG size after the 3rd cycle can be attributed to the destabilization of effective pinning particles to hinder the PAG boundary movement during the re-austenitization.

cond-mat.mtrl-sci↗

arXiv2020

Post-Heat Treatment Design of High-Strength Low-Alloy Steels Processed by Laser Powder Bed Fusion

In this study, a post-heat treatment design for additively manufactured copper-bearing high-strength low-alloy (HSLA)-100 steel is performed by understanding the process-structure-property relationships. Hot isostatic pressing (HIP) is designed to reduce the porosity from 3% to less than 1% for the HSLA-100 steel processed by laser powder bed fusion (LPBF). Quenching dilatometry is employed to design the HIP parameters with the optimized cooling rate for the maximum amount of martensite transformed after HIP. Afterward, a post-heat treatment step with cyclic re-austenitization is introduced for an effective grain refinement to compensate the coarsened microstructure after HIP. Finally, tempering is optimized through microstructure characterization and microhardness. A two-fold increase in the yield strength of the HSLA with tailored microstructure during post-heat treatment is achieved in comparison with the as-built HSLA.

cond-mat.mtrl-sci↗