Vision-based grasping of unknown objects in unstructured environments is a key challenge for autonomous robotic manipulation. A practical grasp synthesis system is required to generate a diverse set of 6-DoF grasps from which a task-relevant grasp can be executed. Although generative models are suitable for learning such complex data distributions, existing models have limitations in grasp quality, long training times, and a lack of flexibility for task-specific generation. In this work, we present GraspLDM- a modular generative framework for 6-DoF grasp synthesis that uses diffusion models as priors in the latent space of a VAE. GraspLDM learns a generative model of object-centric SE (3) grasp poses conditioned on point clouds. GraspLDM’s architecture enables us to train task-specific models efficiently by only re-training a small denoising network in the low-dimensional latent space, as opposed to existing models that need expensive re-training. Our framework provides robust and scalable models on both full and single-view point clouds. GraspLDM models trained with simulation data transfer well to the real world without any further fine-tuning. Our models provide an 80% success rate for 80 grasp attempts of diverse test objects across two real-world robotic setups.

Background: The World Health Organization (WHO) provides protocols for the diagnosis of malaria. One of them is related to the staining process of blood samples to guarantee the correct parasite visualization. Ensuring the quality of the staining procedure on thick blood smears (TBS) is a difficult task, especially in rural centres, where there are factors that can affect the smear quality (e.g. types of reagents employed, place of sample preparation, among others). This work presents an analysis of an image-based approach to evaluate the coloration quality of the staining process of TBS used for malaria diagnosis. Methods: According to the WHO, there are different coloration quality descriptors of smears. Among those, the background colour is one of the best indicators of how well the staining process was conducted. An image database with 420 images (corresponding to 42 TBS samples) was created for analysing and testing image-based algorithms to detect the quality of the coloration of TBS. Background segmentation techniques were explored (based on RGB and HSV colour spaces) to separate the background and foreground (leukocytes, platelets, parasites) information. Then, different features (PCA, correlation, Histograms, variance) were explored as image criteria of coloration quality on the extracted background information; and evaluated according to their capability to classify images as with Good or Bad coloration quality from TBS. Results: For background segmentation, a thresholding-based approach in the SV components of the HSV colour space was selected. It provided robustness separating the background information independently of its coloration quality. On the other hand, as image criteria of coloration quality, among the 19 feature vectors explored, the best one corresponds to the 15-bins histogram of the Hue component with classification rates of > 97%. Conclusions: An analysis of an image-based approach to describe the coloration quality of TBS was presented. It was demonstrated that if a robust background segmentation is conducted, the histogram of the H component from the HSV colour space is the best feature vector to discriminate the coloration quality of the smears. These results are the baseline for automating the estimation of the coloration quality, which has not been studied before, but that can be crucial for automating TBS’s analysis for assisting malaria diagnosis process.

Geological formations, environmental conditions, and soil mechanics frequently generate undesired effects on rovers' mobility, such as slippage or sinkage. Underestimating these undesired effects may compromise the rovers' operation and lead to a premature end of the mission. Minimizing mobility risks becomes a priority for colonising the Moon and Mars. However, addressing this challenge cannot be treated equally for every celestial body since the control strategies may differ; e.g. the low latency Earth-Moon communication allows constant monitoring and controls, something not feasible on Mars. This letter proposes a Hazard Information System (HIS) that estimates the rover's mobility risks (e.g. slippage) using proprioceptive sensors and Machine Learning (supervised and unsupervised). A Graphical User Interface was created to assist human-teleoperation tasks by presenting mobility risk indicators. The system has been developed and evaluated in the lunar analogue facility (LunaLab) at the University of Luxembourg. A real rover and eight participants were part of the experiments. Results demonstrate the benefits of the HIS in the decision-making processes of the operator's response to overcome hazardous situations.