Machine Learning for Robotics researchhttps://srl.cit.tum.de/2026-04-18T21:58:47+00:00FeedCreator 1.8 (info@mypapit.net)Dense Map Representations2022-07-26T16:34:24+00:002022-07-26T16:34:24+00:00https://srl.cit.tum.de/research/densemaprepresentations?rev=1658846064&do=diffAnonymousanonymous@undisclosed.example.comDense Map Representations
Scalable Volumentric TSDF and Occupancy Mapping (SRL at Imperial College)
More recently, we have been exploring alternative map representations, such as octrees encoding Truncated Signed Distance Fields (TSDF) or occupancy values in a volumetric manner — which ideally lend themselves to efficient memory usage, fast access and spatial scalability, while they can be immediately interfaced with robotic motion planning.
Furthermore, the hierarchical structure allows for …Dense Map Representations
Scalable Volumentric TSDF and Occupancy Mapping (SRL at Imperial College)
More recently, we have been exploring alternative map representations, such as octrees encoding Truncated Signed Distance Fields (TSDF) or occupancy values in a volumetric manner — which ideally lend themselves to efficient memory usage, fast access and spatial scalability, while they can be immediately interfaced with robotic motion planning.
Furthermore, the hierarchical structure allows for …Drones2022-07-26T16:35:48+00:002022-07-26T16:35:48+00:00https://srl.cit.tum.de/research/drones?rev=1658846148&do=diffAnonymousanonymous@undisclosed.example.comDrones
Multicopters (SRL Imperial College)
We run OKVIS on our multicopter drones as a basis for fully autonomous operation. We are exploring several Model-Predictive Controllers (MPC) and are working on model-based motion planning for save navigation through free space that is reconstructed by our dense SLAM algorithms.Drones
Multicopters (SRL Imperial College)
We run OKVIS on our multicopter drones as a basis for fully autonomous operation. We are exploring several Model-Predictive Controllers (MPC) and are working on model-based motion planning for save navigation through free space that is reconstructed by our dense SLAM algorithms.People Tracking & Human-Robot Interaction2022-07-26T16:54:19+00:002022-07-26T16:54:19+00:00https://srl.cit.tum.de/research/human?rev=1658847259&do=diffAnonymousanonymous@undisclosed.example.comPeople Tracking & Human-Robot Interaction
For a safe and efficient interactions between mobile robots and human, it is key for the robot to understand the human's behavior, ranging from the human's 3D body pose to its high-level actions. This research area includes questions about good learning representations for human modeling, robustifying model predictions despite the large variety of human shape and multi-modality of human actions, or leveraging contextual information for accurate predicti…People Tracking & Human-Robot Interaction
For a safe and efficient interactions between mobile robots and human, it is key for the robot to understand the human's behavior, ranging from the human's 3D body pose to its high-level actions. This research area includes questions about good learning representations for human modeling, robustifying model predictions despite the large variety of human shape and multi-modality of human actions, or leveraging contextual information for accurate predicti…Machine Learning (Including Deep Learning)2022-07-26T16:35:22+00:002022-07-26T16:35:22+00:00https://srl.cit.tum.de/research/machinelearning?rev=1658846122&do=diffAnonymousanonymous@undisclosed.example.comMachine Learning (Including Deep Learning)
Volumetric Occupancy Mapping With Probabilistic Depth Completion
We have been using Machine Learning, often Deep Learning, extensively as part of SLAM approaches, particularly in mapping. This includes semantic segmentation and instance segmentation, as well as prediction of geometry, e.g. image-based depth prediction and depth completion. Furthermore, understanding uncertainty of predicted quantities allows for robust interfaces with other components…Machine Learning (Including Deep Learning)
Volumetric Occupancy Mapping With Probabilistic Depth Completion
We have been using Machine Learning, often Deep Learning, extensively as part of SLAM approaches, particularly in mapping. This includes semantic segmentation and instance segmentation, as well as prediction of geometry, e.g. image-based depth prediction and depth completion. Furthermore, understanding uncertainty of predicted quantities allows for robust interfaces with other components…Multi-Sensor SLAM2022-07-26T16:34:05+00:002022-07-26T16:34:05+00:00https://srl.cit.tum.de/research/multisensorslam?rev=1658846045&do=diffAnonymousanonymous@undisclosed.example.comMulti-Sensor SLAM
* Chris Choi (SRL Imperial College London)
Keyframe-Based Visual-Inertial Odometry and SLAM Using Nonlinear Optimisation
Here, we fuse inertial measurements with visual measurements: due to the complementary characteristics of these sensing modalities, they have become a popular choice for accurate SLAM in mobile robotics. While historically the problem has been addressed with filtering, advancements in visual estimation suggest that non-linear optimisation offers superior…Multi-Sensor SLAM
* Chris Choi (SRL Imperial College London)
Keyframe-Based Visual-Inertial Odometry and SLAM Using Nonlinear Optimisation
Here, we fuse inertial measurements with visual measurements: due to the complementary characteristics of these sensing modalities, they have become a popular choice for accurate SLAM in mobile robotics. While historically the problem has been addressed with filtering, advancements in visual estimation suggest that non-linear optimisation offers superior…Physical Interaction2022-07-26T16:35:42+00:002022-07-26T16:35:42+00:00https://srl.cit.tum.de/research/physicalinteraction?rev=1658846142&do=diffAnonymousanonymous@undisclosed.example.comPhysical Interaction
Aerial Manipulation (SRL at Imperial College)
Accurate state estimation and detailed knowledge about the environment are crucial when we want robots to get in physical contact with their surroundings. The Smart Robotics Lab has been working on control algorithms towards mobile / aerial manipulation with the example of a drone writing something (accurately) on a whiteboard.Physical Interaction
Aerial Manipulation (SRL at Imperial College)
Accurate state estimation and detailed knowledge about the environment are crucial when we want robots to get in physical contact with their surroundings. The Smart Robotics Lab has been working on control algorithms towards mobile / aerial manipulation with the example of a drone writing something (accurately) on a whiteboard.Robot Navigation2022-07-26T16:35:32+00:002022-07-26T16:35:32+00:00https://srl.cit.tum.de/research/robotnavigation?rev=1658846132&do=diffAnonymousanonymous@undisclosed.example.comRobot Navigation
Exploration of Unknown Environments (SRL at Imperial College)
When a robot is supposed to explore an unknown environment, it is of paramount importance that it only moves into space wherer it will not collide with obstacles; therefore, we need to leverage respective information of free vs. occupied space from the map representations and perform motion planning in a way to efficiently uncover the unknown map while ensuring safety. SRL has been working on respective information…Robot Navigation
Exploration of Unknown Environments (SRL at Imperial College)
When a robot is supposed to explore an unknown environment, it is of paramount importance that it only moves into space wherer it will not collide with obstacles; therefore, we need to leverage respective information of free vs. occupied space from the map representations and perform motion planning in a way to efficiently uncover the unknown map while ensuring safety. SRL has been working on respective information…Semantic, Object-level and Dynamic SLAM2022-07-26T16:35:02+00:002022-07-26T16:35:02+00:00https://srl.cit.tum.de/research/semanicobjectlevelanddynamicslam?rev=1658846102&do=diffAnonymousanonymous@undisclosed.example.comSemantic, Object-level and Dynamic SLAM
Multi-Object and Object-level Dynamic Mapping
In this very much ongoing work, we are exploring segmentation and tracking of (rigid) objects into submaps. On the one hand, the underlying algorithms depend on instance-level semantic segmentation networks, and on the other hand they employ geometric and photometric tracking (also for identification of moving objects), as well as volumetric mapping. Works includeSemantic, Object-level and Dynamic SLAM
Multi-Object and Object-level Dynamic Mapping
In this very much ongoing work, we are exploring segmentation and tracking of (rigid) objects into submaps. On the one hand, the underlying algorithms depend on instance-level semantic segmentation networks, and on the other hand they employ geometric and photometric tracking (also for identification of moving objects), as well as volumetric mapping. Works include