Submitted Successfully!
To reward your contribution, here is a gift for you: A free trial for our video production service.
Thank you for your contribution! You can also upload a video entry or images related to this topic.
Version Summary Created by Modification Content Size Created at Operation
1
Sebastian Huch
 -- 1950 2023-12-29 17:26:06 |
2 format correct
Wendy Huang
Meta information modification 1950 2024-01-05 06:40:29 |

Video Upload Options

Do you have a full video?
Send video materials Upload full video

Confirm

Are you sure to Delete?
Yes No
Cite
If you have any further questions, please contact Encyclopedia Editorial Office.
Huch, S.; Lienkamp, M. LiDAR Local Domain Adaptation for Autonomous Vehicles. Encyclopedia. Available online: https://encyclopedia.pub/entry/53279 (accessed on 20 May 2024).
Huch S, Lienkamp M. LiDAR Local Domain Adaptation for Autonomous Vehicles. Encyclopedia. Available at: https://encyclopedia.pub/entry/53279. Accessed May 20, 2024.
Huch, Sebastian, Markus Lienkamp. "LiDAR Local Domain Adaptation for Autonomous Vehicles" Encyclopedia, https://encyclopedia.pub/entry/53279 (accessed May 20, 2024).
Huch, S., & Lienkamp, M. (2023, December 29). LiDAR Local Domain Adaptation for Autonomous Vehicles. In Encyclopedia. https://encyclopedia.pub/entry/53279
Huch, Sebastian and Markus Lienkamp. "LiDAR Local Domain Adaptation for Autonomous Vehicles." Encyclopedia. Web. 29 December, 2023.
LiDAR Local Domain Adaptation for Autonomous Vehicles
Edit
This entry is adapted from the peer-reviewed paper 10.3390/s23249913

Perception algorithms for autonomous vehicles demand large, labeled datasets. Real-world data acquisition and annotation costs are high, making synthetic data from simulation a cost-effective option. However, training on one source domain and testing on a target domain can cause a domain shift attributed to local structure differences, resulting in a decrease in the model’s performance. Domain adaptation is a form of transfer learning that aims to minimize the domain shift between datasets.

autonomous vehicles domain adaptation domain shift LiDAR object detection

1. Introduction

Autonomous vehicles (AVs) are changing the automotive industry, promising improved road safety and a reduction in emissions [1]. Autonomous shuttles are already operating in restricted areas on public roads, even without safety drivers, but this approach to AVs does not scale well. AVs are equipped with sensor arrays, including cameras, LiDAR sensors, and RaDAR sensors, and are based on data-centric neural networks to perform tasks such as object detection. While training of these object detection networks is usually performed when supervised, the choice and characteristics of the datasets play a crucial role in the network’s real-world performance.
The performance of perception algorithms in autonomous vehicles is related to the quality and quantity of the dataset used for training [2][3][4][5]. This performance limitation is known as the data hunger effect [6]. Specifically, the algorithms’ ability to accurately detect and interpret objects in diverse real-world scenarios is based on the datasets’ representation of these conditions. Factors such as differences in the environmental conditions, types of objects, and sensor inaccuracies in the datasets significantly influence the algorithms’ resilience and ability to adapt to these conditions in the real world. This interdependence highlights the need to select diverse and well-curated datasets to develop reliable perception algorithms for AVs. Thus, the selection of a dataset is a critical factor in the overall success of autonomous vehicle perception.
Pairs of images or point clouds and their corresponding labels are required to train an object detection network. Furthermore, to train these networks effectively, it is necessary to have a large amount of labeled data similar to the conditions that will be encountered when the network is used in the real world. When networks are trained in one domain and then applied to another, for example, when using different datasets from the real world, such as KITTI [7], Waymo [8] or nuScenes [9], a real-to-real domain shift can be observed [10][11][12]. To be successful in inference, there must be a high degree of domain similarity between the source dataset used for training and the target data seen during inference. In conclusion, the scalability of AVs is highly dependent on the datasets needed for the perception algorithms.
Gathering and labeling real-world data is a time-consuming and costly undertaking, as it is a manual process and requires additional efforts for each new sensor configuration. An alternative to real-world data is to employ synthetic data created in 3D simulation environments, such as CARLA [13]. This approach is highly scalable, as the data are automatically labeled in the simulation and further safety-critical scenarios can be recorded in the simulation.
However, the divergence between simulation data and real-world data presents unique challenges. Simulation environments such as CARLA, while highly controlled and reproducible, often lack the complex and unpredictable nature of the real world. Key differences include the representation of environmental conditions, such as varying light and weather conditions, and the dynamic behavior of other road users and pedestrians. Data from simulations typically feature more idealized and consistent conditions, whereas real-world data encompass a wide range of variability, i.e., unexpected sensor noise and the often erratic behavior of humans in the real world. Additionally, the physical properties of objects, such as textures and reflectivity, are often simplified in simulations, leading to a gap in the fidelity of sensor data, particularly for LiDAR and camera sensors. This disparity poses significant challenges in training and validating perception systems, as algorithms developed and tested in simulations may not translate effectively to the complexities and unpredictability encountered in real-world environments.
Similarly to the real-to-real domain shift, a sim-to-real domain shift can also be observed. This occurs when object detection networks are trained with synthetic data and deployed in the real world. The source of the sim-to-real domain shift is a combination of various factors. These include the difference in data distribution, diversity, and lack of realism of the virtual sensor model [14], as it is not an exact replication of the physical sensor and its associated characteristics.
Several studies have been conducted to evaluate or quantify the sim-to-real domain shift, using camera [15][16] or LiDAR [17][18] object detection networks for quantification. The authors train multiple networks on a source dataset and evaluate the trained networks on a target dataset to quantify the domain shift. A metric, such as the mean average precision (mAP), is used to calculate the object detection performance, which can be compared to the object detection performance of a network trained and evaluated on the target dataset. For instance, the study [17] shows that the LiDAR sim-to-real domain shift can be as large as 50%, with real-world trained models achieving more than 70% mAP, while models trained with simulated data consistently reach less than 20% mAP on the real-world test dataset. These studies underscore the significance of the sim-to-real domain shift and its implications for safety-critical applications, such as autonomous vehicles. In other words, despite the advantages of using simulation data, these synthetically generated data cannot be used for real-world applications, and expensive real-world data still need to be collected and annotated.

2. Domain Shift

To assess the efficacy of any domain adaptation technique, it is necessary not only to qualitatively examine the adapted samples, but also to quantify the domain shift using the generated samples in a perception task, such as object detection or segmentation [19]. In order to measure the domain shift, perception networks are trained on both the source and target domains and then evaluated on the target domain. This can be performed for object detection [17] or semantic segmentation [11][12]. Training with the adapted dataset and evaluating on the target domain, one can determine the effectiveness of the domain adaptation method [20][21], which, in turn, determines the degree of reduction in the domain shift. To quantify the domain shift specifically for the sim-to-real setting, a method has been introduced that focuses on the local differences between the simulation and the real world by using a scenario-aligned dataset, providing the same scenes in the real world and the simulation [18]. In this research, this dataset will be used to measure the effectiveness of the presented domain adaptation method. An additional way to evaluate the effectiveness of domain adaptation algorithms is to compare the latent spaces produced by a variational autoencoder (VAE) [22]. In [18], latent spaces are also compared, but instead of using a distinct VAE network, the pretrained perception networks are employed to generate the latent feature vectors. These high-dimensional latent feature vectors can be visualized using methods such as t-distributed stochastic neighbor embedding (t-SNE) [23].

3. Domain Adaptation

In general, domain adaptation is a form of transductive transfer learning [24]. This field of research is characterized by the fact that the source domain labels are available, whereas the target domain labels are not. The domain adaptation process only alters the data before passing them to the perception task, such as an object detection network, and does not modify the perception task itself. Since labeled target data are unavailable, it is also called unsupervised domain adaptation [25]. Domain adaptation can be further categorized into domain-invariant feature learning, normalization statistics, and domain mapping. This categorization originates in the image-based domain adaptation as defined in [19]. In addition to these categories, [25] further introduces the category domain-invariant data representation, another data-driven approach specifically for LiDAR-based domain adaptation. These four approaches to domain adaptation will be reviewed in the following.
Domain-invariant feature learning techniques attempt to make the features produced by a feature extractor independent of the domain, meaning that the features of the source and target domains should have the same distribution. To achieve this, there are two approaches: One is to reduce the divergence of features by employing batch statistics [26] or a cross-model loss from 2D images and 3D point clouds [27]. The other approach is to align the features using a discriminator [28][29].
Normalization statistics techniques employ specific forms of batch normalization layers, such as adaptive batch normalization (AdaBN). The idea is that batch norm statistics can acquire domain knowledge and can be transferred from the source to the target domain. However, ref. [30] highlights that simply relying on AdaBN does not result in satisfactory performance for cross-sensor domain adaptation.
Data preprocessing algorithms that create a domain-invariant representation of the source and target data are known as domain-invariant data representation methods. This representation is then used to train and evaluate a perception network. A two-dimensional projection can be used to convert the source and target data into a domain-independent representation [31][32][33]. Alternatively, a voxelization of the point cloud can be performed before it is fed into the perception network [34][35].
Domain mapping techniques can transform images or point clouds from the source domain to the target domain, which can then be used to train a perception network. The aim of these methods is to modify the source distribution to the target distribution without altering the source semantics, meaning that the source labels remain the same and can be used to train the perception network with pseudo-target samples. Research in this area is scarce and can be divided into two approaches, both of which are based on unaltered generative adversarial network (GAN) techniques created for image-to-image conversion. The first category is to create 2D birds-eye-view image projections from 3D point clouds before passing these into the domain mapping algorithm, such as CycleGAN [36]. For example, refs. [37][38][39] use this approach to adapt synthetic point clouds to the real world. Since the output of this method is a 2D top-view projection, just like the input, a 3D point cloud cannot be recovered. The authors use YOLOv3 [40] for object detection to evaluate their methods using adapted 2D projections. Instead of using 2D projected data, the second approach of domain mapping is to generate front-view images as in [41][42]. Front-view images are a lossless projection of 3D point clouds. Therefore, it is possible to obtain 3D point clouds after translating the domains. Refs. [42][43] investigated the dropout of points from real-world point clouds and applied the same technique to synthetic point clouds. Rather than relying on image-based adversarial domain mapping, ref. [44] proposed aligning the data and class distributions of the source and target domains by means of augmentation techniques and minimizing the Kullback–Leibler divergence between the object classes. Ref. [45] employed a non-adversarial domain mapping technique for real-to-real sensor-to-sensor transfer, with a focus on semantic segmentation. The authors accumulated multiple annotated scans using 3D LiDAR SLAM to create a point cloud map, and then generated semi-synthetic scans from the accumulated map with different sensor parameters as the source sensor. There are methods to map the domains of 3D point clouds beyond the field of autonomous driving, as demonstrated in [46]. This paper introduces an unsupervised 3D domain adaptation technique for objects that aligns the source and target distributions both locally and globally. However, ref. [25] highlighted the scarcity of generative domain mapping methods that adapt 3D point clouds instead of 2D images. For this reason, it aims to close this research gap by proposing a novel domain mapping method. Instead of converting the 3D point clouds into an irreversible 2D bird’s-eye view or front-view representation as in [37][38][39][41][42], the method directly adapts point clouds in their 3D representation, leveraging the local relationships between neighboring points. Moreover, instead of adapting the entire scene point clouds and potentially altering the scene semantics as in previous methods, our method adapts point clouds on an object level and ensures the preservation of scene semantics.

References

  1. Costley, A.; Kunz, C.; Sharma, R.; Gerdes, R. Low Cost, Open-Source Platform to Enable Full-Sized Automated Vehicle Research. IEEE Trans. Intell. Veh. 2021, 6, 3–13.
  2. Goodfellow, I.; Bengio Yoshua, C.A. Deep Learning; MIT Press: Cambridge, MA, USA, 2016.
  3. Wu, X.; Sahoo, D.; Hoi, S.C. Recent advances in deep learning for object detection. Neurocomputing 2020, 396, 39–64.
  4. Feng, D.; Haase-Schutz, C.; Rosenbaum, L.; Hertlein, H.; Glaser, C.; Timm, F.; Wiesbeck, W.; Dietmayer, K. Deep Multi-Modal Object Detection and Semantic Segmentation for Autonomous Driving: Datasets, Methods, and Challenges. IEEE Trans. Intell. Transp. Syst. 2021, 22, 1341–1360.
  5. Fernandes, D.; Silva, A.; Névoa, R.; Simões, C.; Gonzalez, D.; Guevara, M.; Novais, P.; Monteiro, J.; Melo-Pinto, P. Point-cloud based 3D object detection and classification methods for self-driving applications: A survey and taxonomy. Inf. Fusion 2021, 68, 161–191.
  6. Gao, B.; Pan, Y.; Li, C.; Geng, S.; Zhao, H. Are We Hungry for 3D LiDAR Data for Semantic Segmentation? A Survey of Datasets and Methods. IEEE Trans. Intell. Transp. Syst. 2022, 23, 6063–6081.
  7. Geiger, A.; Lenz, P.; Stiller, C.; Urtasun, R. Vision meets robotics: The KITTI dataset. Int. J. Robot. Res. 2013, 32, 1231–1237.
  8. Sun, P.; Kretzschmar, H.; Dotiwalla, X.; Chouard, A.; Patnaik, V.; Tsui, P.; Guo, J.; Zhou, Y.; Chai, Y.; Caine, B.; et al. Scalability in Perception for Autonomous Driving: Waymo Open Dataset. In Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 13–19 June 2020; pp. 2443–2451.
  9. Caesar, H.; Bankiti, V.; Lang, A.H.; Vora, S.; Liong, V.E.; Xu, Q.; Krishnan, A.; Pan, Y.; Baldan, G.; Beijbom, O. nuScenes: A multimodal dataset for autonomous driving. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Long Beach, CA, USA, 15–20 June 2019; pp. 11618–11628.
  10. Ben-David, S.; Blitzer, J.; Crammer, K.; Kulesza, A.; Pereira, F.; Vaughan, J.W. A theory of learning from different domains. Mach. Learn. 2010, 79, 151–175.
  11. Wang, Y.; Chen, X.; You, Y.; Erran, L.; Hariharan, B.; Campbell, M.; Weinberger, K.Q.; Chao, W.L. Train in Germany, Test in The USA: Making 3D Object Detectors Generalize. In Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 13–19 June 2020; pp. 11710–11720.
  12. Tsai, D.; Berrio, J.S.; Shan, M.; Worrall, S.; Nebot, E. See Eye to Eye: A Lidar-Agnostic 3D Detection Framework for Unsupervised Multi-Target Domain Adaptation. IEEE Robot. Autom. Lett. 2022, 7, 7904–7911.
  13. Dosovitskiy, A.; Ros, G.; Codevilla, F.; Lopez, A.; Koltun, V. CARLA: An Open Urban Driving Simulator. In Proceedings of the Conference on Robot Learning, Mountain View, CA, USA, 13–15 November 2017; pp. 1–16.
  14. Hegde, D.; Kilic, V.; Sindagi, V.; Cooper, A.B.; Foster, M.; Patel, V.M. Source-free Unsupervised Domain Adaptation for 3D Object Detection in Adverse Weather. In Proceedings of the IEEE International Conference on Robotics and Automation, London, UK, 29 May–2 June 2023; Volume 2023.
  15. Adam, G.; Chitalia, V.; Simha, N.; Ismail, A.; Kulkarni, S.; Narayan, V.; Schulze, M. Robustness and Deployability of Deep Object Detectors in Autonomous Driving. In Proceedings of the 2019 IEEE Intelligent Transportation Systems Conference, ITSC, Auckland, New Zealand, 27–30 October 2019; pp. 4128–4133.
  16. Nowruzi, F.E.; Kapoor, P.; Kolhatkar, D.; Hassanat, F.A.; Laganiere, R.; Rebut, J. How much real data do we actually need: Analyzing object detection performance using synthetic and real data. In Proceedings of the ICML Workshop on AI for Autonomous Driving, Long Beach, CA, USA, 9–15 June 2019.
  17. Dworak, D.; Ciepiela, F.; Derbisz, J.; Izzat, I.; Komorkiewicz, M.; Wojcik, M. Performance of LiDAR object detection deep learning architectures based on artificially generated point cloud data from CARLA simulator. In Proceedings of the 2019 24th International Conference on Methods and Models in Automation and Robotics, MMAR 2019, Miedzyzdroje, Poland, 26–29 August 2019; pp. 600–605.
  18. Huch, S.; Scalerandi, L.; Rivera, E.; Lienkamp, M. Quantifying the LiDAR Sim-to-Real Domain Shift: A Detailed Investigation Using Object Detectors and Analyzing Point Clouds at Target-Level. IEEE Trans. Intell. Veh. 2023, 8, 2970–2982.
  19. Wilson, G.; Cook, D.J. A Survey of Unsupervised Deep Domain Adaptation. ACM Trans. Intell. Syst. Technol. 2020, 11, 51.
  20. Fang, J.; Zhou, D.; Yan, F.; Zhao, T.; Zhang, F.; Ma, Y.; Wang, L.; Yang, R. Augmented LiDAR Simulator for Autonomous Driving. IEEE Robot. Autom. Lett. 2018, 5, 1931–1938.
  21. Manivasagam, S.; Wang, S.; Wong, K.; Zeng, W.; Sazanovich, M.; Tan, S.; Yang, B.; Ma, W.C.; Urtasun, R. LiDARsim: Realistic LiDAR Simulation by Leveraging the Real World. In Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 13–19 June 2020; pp. 11164–11173.
  22. Hubschneider, C.; Roesler, S.; Zöllner, J.M. Unsupervised Evaluation of Lidar Domain Adaptation. In Proceedings of the 2020 IEEE 23rd International Conference on Intelligent Transportation Systems, ITSC 2020, Rhodes, Greece, 20–23 September 2020.
  23. Van Der Maaten, L.; Hinton, G. Visualizing Data using t-SNE. J. Mach. Learn. Res. 2008, 9, 2579–2605.
  24. Pan, S.J.; Yang, Q. A Survey on Transfer Learning. IEEE Trans. Knowl. Data Eng. 2010, 22, 1345–1359.
  25. Triess, L.T.; Dreissig, M.; Rist, C.B.; Zöllner, J.M. A Survey on Deep Domain Adaptation for LiDAR Perception. In Proceedings of the 2021 IEEE Intelligent Vehicles Symposium Workshops (IV Workshops), Nagoya, Japan, 1–17 July 2021; pp. 350–357.
  26. Wu, B.; Zhou, X.; Zhao, S.; Yue, X.; Keutzer, K. SqueezeSegV2: Improved model structure and unsupervised domain adaptation for road-object segmentation from a LiDAR point cloud. In Proceedings of the IEEE International Conference on Robotics and Automation, Montreal, QC, Canada, 20–24 May 2019; Volume 2019.
  27. Jaritz, M.; Vu, T.H.; De Charette, R.; Wirbel, E.; Perez, P. XMUDA: Cross-Modal Unsupervised Domain Adaptation for 3D Semantic Segmentation. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA, 13–19 June 2020.
  28. Jiang, P.; Saripalli, S. LiDARNet: A Boundary-Aware Domain Adaptation Model for Point Cloud Semantic Segmentation. In Proceedings of the IEEE International Conference on Robotics and Automation, Xi’an, China, 30 May–5 June 2021; Volume 2021.
  29. Wang, Z.; DIng, S.; Li, Y.; Zhao, M.; Roychowdhury, S.; Wallin, A.; Sapiro, G.; Qiu, Q. Range adaptation for 3d object detection in LiDAR. In Proceedings of the 2019 International Conference on Computer Vision Workshop, ICCVW 2019, Seoul, Republic of Korea, 27–28 October 2019.
  30. Rist, C.B.; Enzweiler, M.; Gavrila, D.M. Cross-sensor deep domain adaptation for LiDAR detection and segmentation. In Proceedings of the IEEE Intelligent Vehicles Symposium, Paris, France, 9–12 June 2019; Volume 2019.
  31. Triess, L.T.; Peter, D.; Rist, C.B.; Enzweiler, M.; Zollner, J.M. CNN-based synthesis of realistic high-resolution LiDAR data. In Proceedings of the IEEE Intelligent Vehicles Symposium, Paris, France, 9–12 June 2019; Volume 2019.
  32. Shan, T.; Wang, J.; Chen, F.; Szenher, P.; Englot, B. Simulation-based lidar super-resolution for ground vehicles. Robot. Auton. Syst. 2020, 134, 103647.
  33. Elhadidy, A.; Afifi, M.; Hassoubah, M.; Ali, Y.; Elhelw, M. Improved Semantic Segmentation of Low-Resolution 3D Point Clouds Using Supervised Domain Adaptation. In Proceedings of the 2nd Novel Intelligent and Leading Emerging Sciences Conference, NILES 2020, Giza, Egypt, 24–26 October 2020.
  34. Piewak, F.; Pinggera, P.; Zöllner, M. Analyzing the Cross-Sensor Portability of Neural Network Architectures for LiDAR-based Semantic Labeling. In Proceedings of the 2019 IEEE Intelligent Transportation Systems Conference (ITSC), Auckland, New Zealand, 27–30 October 2019.
  35. Yi, L.; Gong, B.; Funkhouser, T. Complete & Label: A domain adaptation approach to semantic segmentation of LiDAR point clouds. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Nashville, TN, USA, 20–25 June 2021.
  36. Zhu, J.Y.; Park, T.; Isola, P.; Efros, A.A. Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks. In Proceedings of the IEEE International Conference on Computer Vision, Venice, Italy, 22–29 October 2017; Volume 2017.
  37. Saleh, K.; Abobakr, A.; Attia, M.; Iskander, J.; Nahavandi, D.; Hossny, M.; Nahvandi, S. Domain adaptation for vehicle detection from bird’s eye view lidar point cloud data. In Proceedings of the 2019 International Conference on Computer Vision Workshop (ICCVW), Seoul, Republic of Korea, 27–28 October 2019.
  38. Sallab, A.E.; Sobh, I.; Zahran, M.; Essam, N. LiDAR Sensor modeling and Data augmentation with GANs for Autonomous driving. arXiv 2019, arXiv:1905.07290.
  39. Sallab, A.E.; Sobh, I.; Zahran, M.; Shawky, M. Unsupervised Neural Sensor Models for Synthetic LiDAR Data Augmentation. arXiv 2019, arXiv:1911.10575.
  40. Redmon, J.; Farhadi, A. YOLOv3: An Incremental Improvement. arXiv 2018, arXiv:1804.02767.
  41. Caccia, L.; Hoof, H.V.; Courville, A.; Pineau, J. Deep Generative Modeling of LiDAR Data. In Proceedings of the IEEE International Conference on Intelligent Robots and Systems, Macau, China, 3–8 November 2019.
  42. Nakashima, K.; Kurazume, R. Learning to Drop Points for LiDAR Scan Synthesis. In Proceedings of the IEEE International Conference on Intelligent Robots and Systems, Prague, Czech Republic, 27 September–1 October 2021.
  43. Zhao, S.; Wang, Y.; Li, B.; Wu, B.; Gao, Y.; Xu, P.; Darrell, T.; Keutzer, K. ePointDA: An End-to-End Simulation-to-Real Domain Adaptation Framework for LiDAR Point Cloud Segmentation. In Proceedings of the 35th AAAI Conference on Artificial Intelligence, AAAI 2021, Online, 2–9 February 2021; Volume 4B.
  44. Alonso, I.; Riazuelo, L.; Montesano, L.; Murillo, A.C. Domain adaptation in LiDAR semantic segmentation. arXiv 2020, arXiv:2010.12239.
  45. Langer, F.; Milioto, A.; Haag, A.; Behley, J.; Stachniss, C. Domain Transfer for Semantic Segmentation of LiDAR Data using Deep Neural Networks. In Proceedings of the IEEE International Conference on Intelligent Robots and Systems, Las Vegas, NV, USA, 24 October 2020–24 January 2021.
  46. Qin, C.; You, H.; Wang, L.; Jay Kuo, C.C.; Fu, Y. PointDAN: A multi-scale 3D domain adaption network for point cloud representation. In Advances in Neural Information Processing Systems; The MIT Press: Cambridge, MA, USA, 2019; Volume 32.
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.
Upload a video for this entry
Information
Subjects: Computer Science, Artificial Intelligence; Robotics
Contributors MDPI registered users' name will be linked to their SciProfiles pages. To register with us, please refer to https://encyclopedia.pub/register :
Sebastian Huch
,
Markus Lienkamp
View Times: 154
Revisions: 2 times (View History)
Update Date: 05 Jan 2024
Table of Contents
    1000/1000
    Hot Most Recent
    Feedback