Semantic Scene Completion (SSC) aims to perform geometric completion and semantic segmentation simultaneously. Despite the promising results achieved by existing studies, the inherently ill-posed nature of the task presents significant challenges in diverse driving scenarios. This paper introduces TALoS, a novel test-time adaptation approach for SSC that excavates the information available in driving environments. Specifically, we focus on that observations made at a certain moment can serve as Ground Truth (GT) for scene completion at another moment. Given the characteristics of the LiDAR sensor, an observation of an object at a certain location confirms both 1) the occupation of that location and 2) the absence of obstacles along the line of sight from the LiDAR to that point. TALoS utilizes these observations to obtain self-supervision about occupancy and emptiness, guiding the model to adapt to the scene in test time. In a similar manner, we aggregate reliable SSC predictions among multiple moments and leverage them as semantic pseudo-GT for adaptation. Further, to leverage future observations that are not accessible at the current time, we present a dual optimization scheme using the model in which the update is delayed until the future observation is available. Evaluations on the SemanticKITTI validation and test sets demonstrate that TALoS significantly improves the performance of the pre-trained SSC model.

Automatically generating agile whole-body motions for legged and humanoid robots remains a fundamental challenge in robotics. While numerous trajectory optimization approaches have been proposed, there is no clear guideline on how the choice of floating-base space parameterization affects performance, especially for agile behaviors involving complex contact dynamics. In this paper, we present a comparative study of different parameterizations for direct transcription-based trajectory optimization of agile motions in legged systems. We systematically evaluate several common choices under identical optimization settings to ensure a fair comparison. Furthermore, we introduce a novel formulation based on the tangent space of SE(3) for representing the robot's floating-base pose, which, to our knowledge, has not received attention from the literature. This approach enables the use of mature off-the-shelf numerical solvers without requiring specialized manifold optimization techniques. We hope that our experiments and analysis will provide meaningful insights for selecting the appropriate floating-based representation for agile whole-body motion generation.

Semantic Scene Completion (SSC) aims to perform geometric completion and semantic segmentation simultaneously. Despite the promising results achieved by existing studies, the inherently ill-posed nature of the task presents significant challenges in diverse driving scenarios. This paper introduces TALoS, a novel test-time adaptation approach for SSC that excavates the information available in driving environments. Specifically, we focus on that observations made at a certain moment can serve as Ground Truth (GT) for scene completion at another moment. Given the characteristics of the LiDAR sensor, an observation of an object at a certain location confirms both 1) the occupation of that location and 2) the absence of obstacles along the line of sight from the LiDAR to that point.

Graph neural network (GNN) models play a pivotal role in numerous tasks involving graph-related data analysis. Despite their efficacy, similar to other deep learning models, GNNs are susceptible to adversarial attacks. Even minor perturbations in graph data can induce substantial alterations in model predictions. While existing research has explored various adversarial defense techniques for GNNs, the challenge of defending against adversarial attacks on real-world scale graph data remains largely unresolved. On one hand, methods reliant on graph purification and preprocessing tend to excessively emphasize local graph information, leading to sub-optimal defensive outcomes. On the other hand, approaches rooted in graph structure learning entail significant time overheads, rendering them impractical for large-scale graphs. In this paper, we propose a new defense method named Talos, which enhances the global, rather than local, homophily of graphs as a defense. Experiments show that the proposed approach notably outperforms state-of-the-art defense approaches, while imposing little computational overhead.

It was believed that Hephaestus, the Greek god of metallurgy, created and programmed a giant bronze warrior named Talos to protect Crete. Talos is said to be a futuristic cybernetic creature that can think and feel. It was believed that Hephaestus created the creature as part of his project, combining neurological-computer interfaces and living and non-living components into one massive being. The mythology surrounding the creation of the warrior is also said to be the first example of people thinking about the potential of Artificial Intelligence and intelligent robots. Cut to the present, Artificial Intelligence is all around us, and data and algorithms have become more essential to our lives than we can fathom.

We might think that the conception of robots, AI, and automated machines is a modern phenomenon, but, in fact, the idea had already appeared in Western literature nearly 3,000 years ago. Long before Isaac Asimov conceived the Laws of Robotics (1942) and John McCarthy coined the term "Artificial Intelligence" (1995), Ancient Greeks myths were full of stories about intelligent humanoids. The fact that these mythical humanoids meet the criteria of modern definitions on robotics and AI is impressive in itself. But what's even more astonishing is that these old tales can provide us with valuable teachings and insights into our modern discourse on artificial intelligence. Such stories "perpetuated over millennia, are a testament to the persistence of thinking and talking about what it is to be human and what it means to simulate life," historian Adrienne Mayor, writes in her book Gods and Robots: Myths, Machines, and Ancient Dreams of Technology.

Artificial intelligence (AI) might be all the rage today but the philosophical questions it poses -- can machines replace humans, can AI and machines be trusted, and how can AI be used humanely?-- Greek mythology tackled such questions over 2,500 years ago. The myth of Talos tells of a giant bronze man forged by the god Hephaestus. Created to defend the island of Crete from unwanted visitors, Talos's "body had a single vein, which ran all the way from his neck to his ankle, sealed there with either a bronze nail or a thin membrane of skin." The Talos myth could be the first mention of robots in literature, but there are references to mechanical devices used to carry out a particular physical task that occurred around 3000 B.C., according to Stanford's A Robotic History.

Last week we participated in The IEEE-RAS International Conference on Humanoid Robots (Humanoids) as Gold Sponsors. We took part in the workshops'Towards physical-social human-robot interaction,' and'TALOS: Status & Progress', as invited speakers, as well as offering a Virtual Tour of our legged robots including our latest projects, SOLO12 & Kangaroo. The IEEE-RAS International Conference on Humanoid Robots is the internationally recognized prime event of the humanoid robotics community. Established in 2000 and held annually, the Humanoids Conference is a forum for researchers working in the area of humanoid robots including mechatronics, control, perception, planning, learning, human-robot interaction, biomechanics, artificial intelligence, cognition, and neuroscience. Although this year's event took place virtually, PAL Robotics has previously taken part in Humanoids Conferences around the world, including in Toronto, and Beijing in recent years. At the event, we offered a Virtual Tour of all of our legged robots featuring our Humanoids Team: Sai Kishor, Adrià Roig, and Narcis Miguel.

Data is the key factor to drive the development of machine learning (ML) during the past decade. However, high-quality data, in particular labeled data, is often hard and expensive to collect. To leverage large-scale unlabeled data, self-supervised learning, represented by contrastive learning, is introduced. The objective of contrastive learning is to map different views derived from a training sample (e.g., through data augmentation) closer in their representation space, while different views derived from different samples more distant. In this way, a contrastive model learns to generate informative representations for data samples, which are then used to perform downstream ML tasks.

This week, Oxford University announced that American billionaire and philanthropist Stephen A. Schwarzman had gifted the university with its largest cash donation ever--£150 million--to fund (among other things) an institution to investigate the ethics of artificial intelligence. Mr. Schwarzman said that universities need to serve as advisers on the ethics of artificial intelligence and technological advances. While it is certainly true that the technology has moved rapidly ahead of the legislation that patrols it, this is hardly the first time people have thought about the ethics of AI. As any sci-fi buff will tell you, we have been mulling over the ethical ramifications of technologies we didn't possess for a century. What they might not know, however, is that people have been thinking about the potentials and pitfalls of the robot world for thousands of years.