ML Systems Wiki Knowledge base for ML systems, autonomy, and embodied AI
ESC

Tags

239 tags across the wiki

paper 114 autonomous-driving 92 foundation-model 55 transformer 53 vla 49 planning 42 robotics 41 computer-vision 36 perception 32 ilya-30 29 multimodal 29 nlp 26 end-to-end 24 language-modeling 24 llm 17 reasoning 17 imitation-learning 16 3d-occupancy 15 vlm 15 bev 14 diffusion 13 e2e 12 reinforcement-learning 12 world-model 12 chain-of-thought 10 benchmark 9 scaling 9 cross-embodiment 7 driving 6 gaussian-splatting 6 generative-models 6 image-classification 6 information-theory 6 questions 6 self-supervised 6 sources 6 alignment 5 attention 5 cnn 5 foundation 5 knowledge-distillation 5 language-model 5 prediction 5 simulation 5 evaluation 4 image-generation 4 instruction-tuning 4 mixture-of-experts 4 rnn 4 sequence-to-sequence 4 sparse-representation 4 video-prediction 4 explainability 3 flow-matching 3 lstm 3 map 3 occupancy 3 open-source 3 semantic-segmentation 3 sequence-modeling 3 trajectory-prediction 3 vectorized-representation 3 3d-detection 2 3d-perception 2 3d-reconstruction 2 action-representation 2 autonomy 2 autoregressive 2 bimanual 2 closed-loop 2 complexity-theory 2 dataset 2 deployment 2 distributed-training 2 efficient-inference 2 embodied 2 fine-tuning 2 foundation-models 2 foundational 2 gaussian-representation 2 generation 2 generative 2 human-interaction 2 humanoid 2 manipulation 2 memory-augmented-networks 2 ml 2 multi-camera 2 multilingual 2 object-detection 2 parameter-efficient-fine-tuning 2 prompting 2 real-time 2 regularization 2 relational-reasoning 2 residual-networks 2 rlhf 2 scaling-laws 2 segmentation 2 self-improvement 2 self-supervised-learning 2 state-space 2 systems 2 thermodynamics 2 vision-language-model 2 vision-transformer 2 visual-question-answering 2 zero-shot 2 3d 1 3d-scene 1 3d-semantic-occupancy 1 agenda 1 agentic 1 agi 1 algorithmic-information-theory 1 algorithmic-randomness 1 asynchronous 1 attention-mechanism 1 batch 1 bayesian-inference 1 behavior-forecasting 1 camera-fusion 1 classifier-guidance 1 combinatorial-optimization 1 comparison 1 compression 1 computability 1 concept 1 contrastive-learning 1 control 1 convolutional-neural-networks 1 corpus 1 course 1 data-collection 1 decoupled 1 deep-learning 1 denoising 1 depth-estimation 1 dexterous-manipulation 1 differentiable-programming 1 diffusion-policy 1 diffusion-transformer 1 dilated-convolutions 1 dropout 1 efficient 1 embodied-ai 1 embodiment 1 emergent-abilities 1 end-to-end-learning 1 evaluation-metric 1 few-shot 1 few-shot-learning 1 foundations 1 frontend 1 gaussian 1 gaussian-rendering 1 generalist-agent 1 generalization 1 gpu-training 1 graph-neural-networks 1 grounding 1 grpo 1 hierarchical 1 high-frequency-control 1 hosting 1 ilya 1 image-captioning 1 image-text-retrieval 1 in-context-learning 1 inductive-bias 1 intelligence-measurement 1 interactive-annotation 1 interactive-segmentation 1 knowledge-preservation 1 kolmogorov-complexity 1 lanegcn 1 locomotion 1 machine-translation 1 mamba 1 mdl 1 message-passing 1 minimum-description-length 1 model-parallelism 1 model-predictive-control 1 model-selection 1 modular 1 molecular-property-prediction 1 multi-embodiment 1 multi-task 1 natural-language 1 neural-radiance-fields 1 neuro-symbolic 1 obsidian 1 open-world 1 optimization 1 orchestration 1 parallel-architecture 1 parameter-efficient 1 permutation-invariance 1 personalization 1 physical-ai 1 pipeline-parallelism 1 pointer-mechanism 1 privileged-supervision 1 probabilistic-planning 1 proprioception 1 quantization 1 queue 1 radar 1 recurrent-neural-networks 1 representation-learning 1 scene-understanding 1 search 1 seminal 1 sensor-fusion 1 set-modeling 1 siamese-networks 1 simulator 1 source 1 sparse-models 1 spatial-reasoning 1 speech-recognition 1 survey 1 synthesis 1 taxonomy 1 temporal 1 temporal-modeling 1 thesis 1 tokenization 1 tool-use 1 training 1 uniad 1 unified-stack 1 vanishing-gradients 1 variational-autoencoders 1 video-generation 1 video-understanding 1 visual-traces 1 vit 1

Pages tagged bev

BEVDiffuser: Plug-and-Play Diffusion Model for BEV Denoising with Ground-Truth Guidance
source-summary

**[Read on arXiv](https://arxiv.org/abs/2502.19694)** BEVDiffuser addresses a fundamental but under-explored problem in BEV-based perception: the inherent noise in BEV feature maps caused by sensor limitations and the l…
BEVFormer v2: Adapting Modern Image Backbones to Bird's-Eye-View Recognition via Perspective Supervision
paper

📄 **[Read on arXiv](https://arxiv.org/abs/2211.10439)** BEVFormer v2 addresses a critical bottleneck in camera-based 3D perception for autonomous driving: the inability to leverage powerful modern 2D image backbones (e.…
BEVFormer: Learning Bird's-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformers
source-summary

📄 **[Read on arXiv](https://arxiv.org/abs/2203.17270)** Li, Wang, Li, Xie, Sima, Lu, Yu, Dai (Shanghai AI Lab / Nanjing University / HKU), ECCV, 2022. - [Paper](https://arxiv.org/abs/2203.17270) BEVFormer generates a un…
Bevnext Reviving Dense Bev Frameworks For 3D Object Detection
paper

📄 [arXiv:2312.01696](https://arxiv.org/abs/2312.01696) BEVNeXt revives dense BEV (bird's-eye-view) frameworks for camera-based 3D object detection, demonstrating that with the right design choices, dense approaches can…
Drive-OccWorld: Driving in the Occupancy World
source-summary

📄 **[Read on arXiv](https://arxiv.org/abs/2408.14197)** Drive-OccWorld introduces a vision-centric 4D occupancy forecasting world model that directly integrates with end-to-end planning. The core premise is that current…
Fb Bev Bev Representation From Forward Backward View Transformations
paper

📄 **[Read on arXiv](https://arxiv.org/abs/2308.02236)** FB-BEV addresses a fundamental tension in camera-based BEV perception for autonomous driving: **forward projection** methods (like Lift-Splat-Shoot) generate BEV f…
FlashOcc: Fast and Memory-Efficient Occupancy Prediction via Channel-to-Height Plugin
paper

📄 **[Read on arXiv](https://arxiv.org/abs/2311.12058)** Occupancy prediction has emerged as a powerful perception paradigm for autonomous driving, predicting per-voxel semantic labels in 3D space to handle arbitrary obj…
GaussianBeV: 3D Gaussian Representation meets Perception Models for BeV Segmentation
paper

📄 **[Read on arXiv](https://arxiv.org/abs/2407.14108)** Bird's-eye view (BEV) semantic segmentation from multi-camera images is a core perception task in autonomous driving, but existing image-to-BEV transformation meth…
GaussianLSS: Toward Real-world BEV Perception with Depth Uncertainty via Gaussian Splatting
source-summary

📄 **[Read on arXiv](https://arxiv.org/abs/2504.01957)** Bird's-Eye View (BEV) perception faces a fundamental trade-off between accuracy and computational efficiency. High-performing 3D projection methods like BEVFormer…
Lift Splat Shoot Encoding Images From Arbitrary Camera Rigs By Implicitly Unprojecting To 3D
source-summary

📄 **[Read on arXiv](https://arxiv.org/abs/2008.05711)** Lift, Splat, Shoot (LSS) introduced a differentiable pipeline for transforming multi-camera images into a unified bird's-eye view (BEV) representation without requ…
Open Questions: BEV Perception & 3D Occupancy
query

Stream-specific open questions for the BEV perception and 3D occupancy pillar. See wiki/queries/open-questions for the full tree across all streams. 1. **Dense vs. sparse vs. Gaussian:** BEVNeXt revived dense BEV to 64.…
SelfOcc: Self-Supervised Vision-Based 3D Occupancy Prediction
paper

📄 **[Read on arXiv](https://arxiv.org/abs/2311.12754)** SelfOcc (Huang et al., Tsinghua University, CVPR 2024) introduces the first self-supervised framework for vision-based 3D occupancy prediction that works with mult…
VLP: Vision Language Planning for Autonomous Driving
source-summary

📄 **[Read on arXiv](https://arxiv.org/abs/2401.05577)** VLP (Vision Language Planning) by Pan et al. (CVPR 2024) represents a fundamentally different approach to using language in autonomous driving compared to instruct…
WoTE: End-to-End Driving with Online Trajectory Evaluation via BEV World Model
source-summary

📄 **[Read on arXiv](https://arxiv.org/abs/2504.01941)** End-to-end driving models typically output a single trajectory and trust it entirely, with no mechanism to evaluate whether the predicted path is safe before execu…