Modeling spatial-temporal interactions among neighboring agents is at the heart of multi-agent problems such as motion forecasting and crowd navigation. Despite notable progress, it remains unclear to which extent modern representations can capture the causal relationships behind agent interactions. In this work, we take an in-depth look at the causal awareness of these representations, from computational formalism to real-world practice. First, we revisit the notion of non-causal robustness studied in the recent CausalAgents benchmark [56]. We show that existing representations are already partially resilient to perturbations of non-causal agents, and yet modeling indirect causal effects involving mediator agents remains challenging. To address this challenge, we introduce a metric learning approach that regularizes latent representations with causal annotations. Our controlled experiments show that this approach not only leads to higher degrees of causal awareness but also yields stronger out-of-distribution robustness. To further operationalize it in practice, we propose a sim-to-real causal transfer method via cross-domain multi-task learning. Experiments on trajectory prediction datasets show that our method can significantly boost generalization, even in the absence of real-world causal annotations, where we acquire higher prediction accuracy by only using 25% of real-world data. We hope our work provides a new perspective on the challenges and potential pathways toward causally-aware representations of multi-agent interactions. Our code is available at https://github.com/vita-epfl/CausalSim2Real.

Conventional human trajectory prediction models rely on clean curated data, requiring specialized equipment or manual labeling, which is often impractical for robotic applications. The existing predictors tend to overfit to clean observation affecting their robustness when used with noisy inputs. In this work, we propose MonoTransmotion (MT), a Transformerbased framework that uses only a monocular camera to jointly solve localization and prediction tasks. Our framework has two main modules: Bird's Eye View (BEV) localization and trajectory prediction. The BEV localization module estimates the position of a person using 2D human poses, enhanced by a novel directional loss for smoother sequential localizations. The trajectory prediction module predicts future motion from these estimates. We show that by jointly training both tasks with our unified framework, our method is more robust in real-world scenarios made of noisy inputs. We validate our MT network on both curated and non-curated datasets. On the curated dataset, MT achieves around 12% improvement over baseline models on BEV localization and trajectory prediction. On real-world non-curated dataset, experimental results indicate that MT maintains similar performance levels, highlighting its robustness and generalization capability. The code is available at https://github.com/vita-epfl/MonoTransmotion.

Premature convergence is a critical issue of Particle Swarm Optimization (PSO). The weak global search capability causes particles trapped in local minima at early stage of learning process. There is several research dedicate to solve this problem over the decade. This paper proposes a new algorithm combined Isolation Forest and Particle Swarm Optimization called Isolation Forest Backward Particle Swarm Optimization (IFB-PSO). The proposed new learning scheme helps particles escaping from local minima. The particle will jump backward to the targeted position when the particle trapped over specific iterations. The destination is precisely selected by Isolation Forest to endow the backward particle hopeful future. IFB-PSO is evaluated by a classic benchmark suite, cart-pole problem, and mountain car problem. Experimental results show that IFB-PSO gets competitive results on the benchmark suite with different dimensions and two control problems in comparison with 11 well-known optimization algorithms. The behavior of backward particles is also analyzed to inspect the utility and efficiency of the backward process.

In this article, the adaptive-network-based fuzzy inference system (ANFIS) is combined with the double deepQ-network (DDQN) to realize a fuzzy DDQN (FDDQN) such that a humanoid robot can generate a linear inverted pendulum model-based gait pattern in real time. The FDDQN not only allows the humanoid robot to correct the gait pattern instantly but also improves its stability. The proposed scheme is designed and implemented in a toddler-sized humanoid robot called Louis. First, four pressure sensors are installed on the bottom of the sole and one inertial measurement unit is set up on the trunk of the robot. A wireless communication chip is employed to transfer the data to a computer to determine the required parameters for the robot. Next, a control system based on the Linux operating system is developed. The values of the center of pressure and acceleration obtained with the ANFIS are adopted to train the DDQN. The proposed neural network comprises four layers, and the model is cautiously selected to avoid overfitting. The proposed scheme is verified using a robot simulator and then real-time-tested on Louis. The experimental results indicate that the FDDQN can provide the robot timely feedback during walking as well as helps it in adjusting the gait pattern independently. The balancing of the robot through effective dynamic feedback is similar to the balancing ability of an infant learning to walk.