MineRL Diamond 2021 Competition: Overview, Results, and Lessons Learned

Reinforcement learning competitions advance the field by providing appropriate scope and support to develop solutions toward a specific problem. To promote the development of more broadly applicable methods, organizers need to enforce the use of general techniques, the use of sample-efficient methods, and the reproducibility of the results. While beneficial for the research community, these restrictions come at a cost -- increased difficulty. If the barrier for entry is too high, many potential participants are demoralized. With this in mind, we hosted the third edition of the MineRL ObtainDiamond competition, MineRL Diamond 2021, with a separate track in which we permitted any solution to promote the participation of newcomers. With this track and more extensive tutorials and support, we saw an increased number of submissions. The participants of this easier track were able to obtain a diamond, and the participants of the harder track progressed the generalizable solutions in the same task.

Multi-Modal Fusion in Contact-Rich Precise Tasks via Hierarchical Policy Learning

Combined visual and force feedback play an essential role in contact-rich robotic manipulation tasks. Current methods focus on developing the feedback control around a single modality while underrating the synergy of the sensors. Fusing different sensor modalities is necessary but remains challenging. A key challenge is to achieve an effective multi-modal and generalized control scheme to novel objects with precision. This paper proposes a practical multi-modal sensor fusion mechanism using hierarchical policy learning. To begin with, we use a self-supervised encoder that extracts multi-view visual features and a hybrid motion/force controller that regulates force behaviors. Next, the multi-modality fusion is simplified by hierarchical integration of the vision, force, and proprioceptive data in the reinforcement learning (RL) algorithm. Moreover, with hierarchical policy learning, the control scheme can exploit the visual feedback limits and explore the contribution of individual modality in precise tasks. Experiments indicate that robots with the control scheme could assemble objects with 0.25mm clearance in simulation. The system could be generalized to widely varied initial configurations and new shapes. Experiments validate that the simulated system can be robustly transferred to reality without fine-tuning.

Video-driven Neural Physically-based Facial Asset for Production

Production-level workflows for producing convincing 3D dynamic human faces have long relied on a disarray of labor-intensive tools for geometry and texture generation, motion capture and rigging, and expression synthesis. Recent neural approaches automate individual components but the corresponding latent representations cannot provide artists explicit controls as in conventional tools. In this paper, we present a new learning-based, video-driven approach for generating dynamic facial geometries with high-quality physically-based assets. Two key components are well-structured latent spaces due to dense temporal samplings from videos and explicit facial expression controls to regulate the latent spaces. For data collection, we construct a hybrid multiview-photometric capture stage, coupling with an ultra-fast video camera to obtain raw 3D facial assets. We then model the facial expression, geometry and physically-based textures using separate VAEs with a global MLP-based expression mapping across the latent spaces, to preserve characteristics across respective attributes while maintaining explicit controls over geometry and texture. We also introduce to model the delta information as wrinkle maps for physically-base textures, achieving high-quality rendering of dynamic textures. We demonstrate our approach in high-fidelity performer-specific facial capture and cross-identity facial motion retargeting. In addition, our neural asset along with fast adaptation schemes can also be deployed to handle in-the-wild videos. Besides, we motivate the utility of our explicit facial disentangle strategy by providing promising physically-based editing results like geometry and material editing or winkle transfer with high realism. Comprehensive experiments show that our technique provides higher accuracy and visual fidelity than previous video-driven facial reconstruction and animation methods.

NeuVV: Neural Volumetric Videos with Immersive Rendering and Editing

Some of the most exciting experiences that Metaverse promises to offer, for instance, live interactions with virtual characters in virtual environments, require real-time photo-realistic rendering. 3D reconstruction approaches to rendering, active or passive, still require extensive cleanup work to fix the meshes or point clouds. In this paper, we present a neural volumography technique called neural volumetric video or NeuVV to support immersive, interactive, and spatial-temporal rendering of volumetric video contents with photo-realism and in real-time. The core of NeuVV is to efficiently encode a dynamic neural radiance field (NeRF) into renderable and editable primitives. We introduce two types of factorization schemes: a hyper-spherical harmonics (HH) decomposition for modeling smooth color variations over space and time and a learnable basis representation for modeling abrupt density and color changes caused by motion. NeuVV factorization can be integrated into a Video Octree (VOctree) analogous to PlenOctree to significantly accelerate training while reducing memory overhead. Real-time NeuVV rendering further enables a class of immersive content editing tools. Specifically, NeuVV treats each VOctree as a primitive and implements volume-based depth ordering and alpha blending to realize spatial-temporal compositions for content re-purposing. For example, we demonstrate positioning varied manifestations of the same performance at different 3D locations with different timing, adjusting color/texture of the performer's clothing, casting spotlight shadows and synthesizing distance falloff lighting, etc, all at an interactive speed. We further develop a hybrid neural-rasterization rendering framework to support consumer-level VR headsets so that the aforementioned volumetric video viewing and editing, for the first time, can be conducted immersively in virtual 3D space.

EREBA: Black-box Energy Testing of Adaptive Neural Networks

Recently, various Deep Neural Network (DNN) models have been proposed for environments like embedded systems with stringent energy constraints. The fundamental problem of determining the robustness of a DNN with respect to its energy consumption (energy robustness) is relatively unexplored compared to accuracy-based robustness. This work investigates the energy robustness of Adaptive Neural Networks (AdNNs), a type of energy-saving DNNs proposed for many energy-sensitive domains and have recently gained traction. We propose EREBA, the first black-box testing method for determining the energy robustness of an AdNN. EREBA explores and infers the relationship between inputs and the energy consumption of AdNNs to generate energy surging samples. Extensive implementation and evaluation using three state-of-the-art AdNNs demonstrate that test inputs generated by EREBA could degrade the performance of the system substantially. The test inputs generated by EREBA can increase the energy consumption of AdNNs by 2,000% compared to the original inputs. Our results also show that test inputs generated via EREBA are valuable in detecting energy surging inputs.

Artemis: Articulated Neural Pets with Appearance and Motion synthesis

We human are entering into a virtual era, and surely want to bring animals to virtual world as well for companion. Yet, computer-generated (CGI) furry animals is limited by tedious off-line rendering, let alone interactive motion control. In this paper, we present ARTEMIS, a novel neural modeling and rendering pipeline for generating ARTiculated neural pets with appEarance and Motion synthesIS. Our ARTEMIS enables interactive motion control, real-time animation and photo-realistic rendering of furry animals. The core of ARTEMIS is a neural-generated (NGI) animal engine, which adopts an efficient octree based representation for animal animation and fur rendering. The animation then becomes equivalent to voxel level skeleton based deformation. We further use a fast octree indexing, an efficient volumetric rendering scheme to generate appearance and density features maps. Finally, we propose a novel shading network to generate high-fidelity details of appearance and opacity under novel poses. For the motion control module in ARTEMIS, we combine state-of-the-art animal motion capture approach with neural character control scheme. We introduce an effective optimization scheme to reconstruct skeletal motion of real animals captured by a multi-view RGB and Vicon camera array. We feed the captured motion into a neural character control scheme to generate abstract control signals with motion styles. We further integrate ARTEMIS into existing engines that support VR headsets, providing an unprecedented immersive experience where a user can intimately interact with a variety of virtual animals with vivid movements and photo-realistic appearance. Extensive experiments and showcases demonstrate the effectiveness of our ARTEMIS system to achieve highly realistic rendering of NGI animals in real-time, providing daily immersive and interactive experience with digital animals unseen before.

Learning Free Gait Transition for Quadruped Robots via Phase-Guided Controller

Gaits and transitions are key components in legged locomotion. For legged robots, describing and reproducing gaits as well as transitions remain longstanding challenges. Reinforcement learning has become a powerful tool to formulate controllers for legged robots. Learning multiple gaits and transitions, nevertheless, is related to the multi-task learning problems. In this work, we present a novel framework for training a simple control policy for a quadruped robot to locomote in various gaits. Four independent phases are used as the interface between the gait generator and the control policy, which characterizes the movement of four feet. Guided by the phases, the quadruped robot is able to locomote according to the generated gaits, such as walk, trot, pacing and bounding, and to make transitions among those gaits. More general phases can be used to generate complex gaits, such as mixed rhythmic dancing. With the control policy, the Black Panther robot, a medium-dog-sized quadruped robot, can perform all learned motor skills while following the velocity commands smoothly and robustly in natural environment.

HumanNeRF: Generalizable Neural Human Radiance Field from Sparse Inputs

Recent neural human representations can produce high-quality multi-view rendering but require using dense multi-view inputs and costly training. They are hence largely limited to static models as training each frame is infeasible. We present HumanNeRF - a generalizable neural representation - for high-fidelity free-view synthesis of dynamic humans. Analogous to how IBRNet assists NeRF by avoiding per-scene training, HumanNeRF employs an aggregated pixel-alignment feature across multi-view inputs along with a pose embedded non-rigid deformation field for tackling dynamic motions. The raw HumanNeRF can already produce reasonable rendering on sparse video inputs of unseen subjects and camera settings. To further improve the rendering quality, we augment our solution with an appearance blending module for combining the benefits of both neural volumetric rendering and neural texture blending. Extensive experiments on various multi-view dynamic human datasets demonstrate the generalizability and effectiveness of our approach in synthesizing photo-realistic free-view humans under challenging motions and with very sparse camera view inputs.

JueWu-MC: Playing Minecraft with Sample-efficient Hierarchical Reinforcement Learning

Learning rational behaviors in open-world games like Minecraft remains to be challenging for Reinforcement Learning (RL) research due to the compound challenge of partial observability, high-dimensional visual perception and delayed reward. To address this, we propose JueWu-MC, a sample-efficient hierarchical RL approach equipped with representation learning and imitation learning to deal with perception and exploration. Specifically, our approach includes two levels of hierarchy, where the high-level controller learns a policy to control over options and the low-level workers learn to solve each sub-task. To boost the learning of sub-tasks, we propose a combination of techniques including 1) action-aware representation learning which captures underlying relations between action and representation, 2) discriminator-based self-imitation learning for efficient exploration, and 3) ensemble behavior cloning with consistency filtering for policy robustness. Extensive experiments show that JueWu-MC significantly improves sample efficiency and outperforms a set of baselines by a large margin. Notably, we won the championship of the NeurIPS MineRL 2021 research competition and achieved the highest performance score ever.