The goal is to build an autonomous system capable of navigating areas(small scale) without the use of GPS technology. The inspiration of the project is taken from recent advances in Semantic Segmentation and Deep Reinforcement Learning which have proven to be very successful in the context of Autonomous Navigation in the case of cars. We aim to extend it to drones. Navigation of areas involves obstacle avoidance, decision based system to classify input camera feed for commands for navigation. We are currently using ROS for interfacing. We might switch to flytOS depending on it's capabilities which we still have to assess.
- Order of materials required for drone and initial research
- Implementation of initial model (Nvidia TrailNet): https://arxiv.org/abs/1705.02550
- Dec '18: Research on Semantic Segmentation and Deep Reinforcement Learning
- Jan '19: Implementation of models
- Feb '19: Quark presentation
- Mask RCNN: https://arxiv.org/pdf/1703.06870.pdf
- Full Resolution Residual Networks: https://arxiv.org/pdf/1611.08323.pdf
- Pyramid Scene Parsing Network: https://arxiv.org/pdf/1612.01105.pdf
- The One Hundred Layer Tiramisu: https://arxiv.org/pdf/1611.09326.pdf
- Rethinking Atrous Convolution (DeepLabV3): https://arxiv.org/pdf/1706.05587.pdf
- Multi-Path Refinement Networks: https://arxiv.org/pdf/1611.06612.pdf
- Large Kernel Matters: https://arxiv.org/pdf/1703.02719.pdf
- DeepLabV3+: https://arxiv.org/pdf/1802.02611.pdf
- Image Cascade Network (ICNet): https://arxiv.org/pdf/1704.08545.pdf
- Discriminative Feature Network (DFN): http://openaccess.thecvf.com/content_cvpr_2018/papers/Yu_Learning_a_Discriminative_CVPR_2018_paper.pdf
- DenseASPP: http://openaccess.thecvf.com/content_cvpr_2018/papers/Yang_DenseASPP_for_Semantic_CVPR_2018_paper.pdf
- Context Encoding: http://openaccess.thecvf.com/content_cvpr_2018/papers/Zhang_Context_Encoding_for_CVPR_2018_paper.pdf
- Dense Decoder Shortcut Connection: http://openaccess.thecvf.com/content_cvpr_2018/papers/Bilinski_Dense_Decoder_Shortcut_CVPR_2018_paper.pdf
- Bilateral Segmentation Network: https://arxiv.org/pdf/1808.00897v1.pdf
- ExFuse: http://openaccess.thecvf.com/content_ECCV_2018/papers/Zhenli_Zhang_ExFuse_Enhancing_Feature_ECCV_2018_paper.pdf
- Memory-based Deep Reinforcement Learning for Obstacle Avoidance in UAV with Limited Environment Knowledge: https://arxiv.org/abs/1811.03307
Collaborators: