Field-of-View (FoV) adaptive streaming significantly reduces bandwidth requirement of immersive point cloud video (PCV) by only transmitting visible points inside a viewer's FoV. The traditional approaches often focus on trajectory-based 6 degree-of-freedom (6DoF) FoV predictions. The predicted FoV is then used to calculate point visibility. Such approaches do not explicitly consider video content's impact on viewer attention, and the conversion from FoV to point visibility is often error-prone and time-consuming. We reformulate the PCV FoV prediction problem from the cell visibility perspective, allowing for precise decision-making regarding the transmission of 3D data at the cell level based on the predicted visibility distribution. We develop a novel spatial visibility and object-aware graph model (CellSight) that leverages the historical 3D visibility data and incorporates spatial perception, occlusion between points, and neighboring cell correlation to predict the cell visibility in the future. We focus on multi-second ahead prediction to enable the use of long pre-fetching buffers in on-demand streaming, critical for enhancing the robustness to network bandwidth fluctuations. CellSight significantly improves the long-term cell visibility prediction, reducing the prediction Mean Squared Error (MSE) loss by up to 50% compared to the state-of-the-art models when predicting 2 to 5 seconds ahead, while maintaining real-time performance (more than 30fps) for point cloud videos with over 1 million points.

Many XR applications require the delivery of volumetric video to users. Point Cloud has become a popular volumetric video format. A dense point cloud consumes much higher bandwidth than a 2D/360$^{\circ }$video frame. User Field of View (FoV) is more dynamic with 6-DoF movement than 3-DoF movement. To save bandwidth, FoV-adaptive streaming predicts a user's FoV and only downloads point cloud data falling in the predicted FoV. However, it is vulnerable to FoV prediction errors, which can be significant when a long buffer is utilized for smoothed streaming. In this work, we propose a multi-round progressive refinement framework for point cloud video streaming. Instead of sequentially downloading point cloud frames, our solution simultaneously downloads/patches multiple frames falling into a sliding time-window, leveraging the inherent scalability of octree-based point-cloud coding. The optimal rate allocation among all tiles of active frames are solved numerically using the heterogeneous tile rate-quality functions calibrated by the predicted user FoV. Multi-frame downloading/patching simultaneously takes advantage of the streaming smoothness resulting from long buffer and the FoV prediction accuracy at short buffer length. We evaluate our streaming solution using simulations driven by real point cloud videos, real bandwidth traces, and 6-DoF FoV traces of real users. Our solution is robust against the bandwidth/FoV prediction errors, and can deliver high and smooth view quality in the face of bandwidth variations and dynamic user and point cloud movements.

Streaming of live 360-degree video allows users to follow a live event from any view point and has already been deployed on some commercial platforms. However, the current systems can only stream the video at relatively low-quality because the entire 360-degree video is delivered to the users under limited bandwidth. Streaming video falling into user field of view (FoV) can improve bandwidth efficiency of 360-degree video delivery. In this paper, we propose to use the idea of “flocking” to simultaneously improve the accuracy of user FoV prediction and video delivery efficiency for live 360-degree video streaming. By assigning variable playback latencies to users in a streaming session based on their network conditions, a “streaming flock” is formed and led by “strong” users with low playback latencies in the front of the flock. We propose a long short-term memory (LSTM) based collaborative FoV prediction scheme where the FoV traces of users in the front of the flock are utilized to predict the FoV of users behind them. Given a predicted FoV, we develop an optimal rate allocation strategy to maximize the perceptual quality. By conducting experiments using real-world user FoV traces and LTE/5 G network bandwidth traces, we evaluate the gains of the proposed strategies over several benchmarks. Our experimental results demonstrate that the proposed streaming system can increase the overall quality dramatically by about 10 dB compared with heuristic FoV prediction strategy. In addition, the network-aware flocking formation can further reduce the video freeze without influencing video quality.

Edge caching will play a critical role in facilitating the emerging content-rich applications. However, it faces many new challenges, in particular, the highly dynamic content popularity and the heterogeneous caching configurations. In this paper, we propose Cocktail Edge Caching, that tackles the dynamic popularity and heterogeneity through ensemble learning. Instead of trying to find a single dominating caching policy for all the caching scenarios, we employ an ensemble of constituent caching policies and adaptively select the best-performing policy to control the cache. Towards this goal, we first show through formal analysis and experiments that different variations of the LFU and LRU policies have complementary performance in different caching scenarios. We further develop a novel caching algorithm that enhances LFU/LRU with deep recurrent neural network (LSTM) based time-series analysis. Finally, we develop a deep reinforcement learning agent that adaptively combines base caching policies according to their virtual hit ratios on parallel virtual caches. Through extensive experiments driven by real content requests from two large video streaming platforms, we demonstrate that CEC not only consistently outperforms all single policies, but also improves the robustness of them. CEC can be well generalized to different caching scenarios with low computation overheads for deployment.

Robot locomotion is a major challenge in robotics. Model-based approaches are vulnerable to model errors, and incur high computation overhead resulted from long control horizon. Model-free approaches are trained with a large number of training samples, which are expensive to obtain. In this paper, we develop a hybrid control and learning framework, called Reinforced iLQR (RiLQR), which combines the advantages of model-based iLQR control with model-free RL policy learning to simultaneously achieve high sample efficiency, low computation overhead, and high robustness against model errors in robot locomotion. Through extensive evaluation on the Mujoco platform, we demonstrate that RiLQR outperforms the state-of-the-art model-based and model-free baselines by big margins in a set of tasks with different complexities.

Edge caching will play a critical role in facilitating the emerging content-rich applications. However, it faces many new challenges, in particular, the highly dynamic content popularity and the heterogeneous caching configurations. In this paper, we propose Cocktail Edge Caching, that tackles the dynamic popularity and heterogeneity through ensemble learning. Instead of trying to find a single dominating caching policy for all the caching scenarios, we employ an ensemble of constituent caching policies and adaptively select the best-performing policy to control the cache. Towards this goal, we first show through formal analysis and experiments that different variations of the LFU and LRU polices have complementary performance in different caching scenarios. We further develop a novel caching algorithm that enhances LFU/LRU with deep recurrent neural network (LSTM) based time-series analysis. Finally, we develop a deep reinforcement learning agent that adaptively combines base caching policies according to their virtual hit ratios on parallel virtual caches. Through extensive experiments driven by real content requests from two large video streaming platforms, we demonstrate that CEC not only consistently outperforms all single policies, but also improves the robustness of them. CEC can be well generalized to different caching scenarios with low computation overheads for deployment.

It is highly challenging to simultaneously achieve high-rate and low-latency in live video streaming. Chunk-based streaming and playback speed adaptation are two promising new trends to achieve high user Quality-of-Experience (QoE). To thoroughly understand their potentials, we develop a detailed chunk-level dynamic model that characterizes how video rate and playback speed jointly control the evolution of a live streaming session. Leveraging on the model, we first study the optimal joint video rate-playback speed adaptation as a non-linear optimal control problem. We further develop model-free joint adaptation strategies using deep reinforcement learning. Through extensive experiments, we demonstrate that our proposed joint adaptation algorithms significantly outperform rate-only adaptation algorithms and the recently proposed low-latency video streaming algorithms that separately adapt video rate and playback speed without joint optimization. In a wide-range of network conditions, the model-based and model-free algorithms can achieve close-to-optimal trade-offs tailored for users with different QoE preferences.

Low-latency is a critical user Quality-of-Experience (QoE) metric for live video streaming. It poses significant challenges for streaming over the Internet. In this paper, we explore the design space of low-latency live streaming by developing dynamic models and optimal adaptation strategies to establish QoE upper bounds as a function of the allowable end-to-end latency. We further develop practical live streaming algorithms within the iterative Linear Quadratic Regulator (iLQR) based Model Predictive Control and Deep Reinforcement Learning frameworks, namely MPC-Live and DRL-Live, to maximize user live streaming QoE by adapting the video bitrate while maintaining low end-to-end video latency in dynamic network environment. Through extensive experiments driven by real network traces, we demonstrate that our live streaming algorithms can achieve close-to-optimal performance within the latency range of two to five seconds.

Streaming of live 360-degree video allows users to follow a live event from any view point and has already been deployed on some commercial platforms. However, the current systems can only stream the video at relatively low-quality because the entire 360-degree video is delivered to the users under limited bandwidth. In this paper, we propose to use the idea of "flocking" to improve the performance of both prediction of field of view (FoV) and caching on the edge servers for live 360-degree video streaming. By assigning variable playback latencies to all the users in a streaming session, a "streaming flock" is formed and led by low latency users in the front of the flock. We propose a collaborative FoV prediction scheme where the actual FoV information of users in the front of the flock are utilized to predict of users behind them. We further propose a network condition aware flocking strategy to reduce the video freeze and increase the chance for collaborative FoV prediction on all users. Flocking also facilitates caching as video tiles downloaded by the front users can be cached by an edge server to serve the users at the back of the flock, thereby reducing the traffic in the core network. We propose a latency-FoV based caching strategy and investigate the potential gain of applying transcoding on the edge server. We conduct experiments using real-world user FoV traces and WiGig network bandwidth traces to evaluate the gains of the proposed strategies over benchmarks. Our experimental results demonstrate that the proposed streaming system can roughly double the effective video rate, which is the video rate inside a user's actual FoV, compared to the prediction only based on the user's own past FoV trajectory, while reducing video freeze. Furthermore, edge caching can reduce the traffic in the core network by about 80%, which can be increased to 90% with transcoding on edge server.

Low-latency is a critical user Quality-of-Experience (QoE) metric for live video streaming. It poses significant challenges for streaming over the Internet. In this paper, we explore the design space of low-latency live video streaming by developing dynamic models and optimal control strategies. We further develop practical live video streaming algorithms within the Model Predictive Control (MPC) framework, namely MPC-Live, to maximize user QoE by adapting the video bitrate while maintaining low end-to-end video latency in dynamic network environment. Through extensive experiments driven by real network traces, we demonstrate that our live video streaming algorithms can improve the performance dramatically within latency range of two to five seconds.