Geospatial machine learning is of growing importance in various global remote-sensing applications, particularly in the realm of vegetation monitoring. However, acquiring accurate ground truth data for geospatial tasks remains a significant challenge, often entailing considerable time and effort. Foundation models, emphasizing pre-training on large-scale data and fine-tuning, show promise but face limitations when applied to geospatial data due to domain differences. Our paper introduces a novel image translation method, combining geospatial-specific pre-training with training and test-time data augmentation. In a case study involving the translation of normalized difference vegetation index (NDVI) values from synthetic aperture radar (SAR) images of cabbage farms, our approach outperformed competitors by 31% in a public competition. It also exceeded the average of the top five teams by 44%. We publish both our image translation method with baseline methods and the geospatial-specific dataset at https://github.com/IBM/SAR2NDVI.

High-resolution flood modeling is enabled by utilizing high-resolution input derived by remote sensing technologies such as Light Detection and Ranging (LiDAR) systems. However, there is a long-standing trade-off between the computational time and spatial resolution for a flood simulation. In this paper, we propose a novel deep learning-based geospatial encoder-decoder for flood modeling consisting of (i) accuracy-preserving coarse-graining of the input topography, (ii) simulating flood with the coarser model, and (iii) downscaling the simulated flood to super-resolution. Our experiments show that our approach accelerates flood simulation up to 50 times faster with 1/16 scale while MSE of 0.0179, which is 10.3% less than the baseline with bilinear interpolation. Especially, we observe 20.5% reduction of MSE on average for the 5% worst cases.

As climate change is increasing the frequency and intensity of climate and weather hazards, improving detection and monitoring of flood events is a priority. Being weather independent and high resolution, Sentinel 1 (S1) radar satellite imagery data has become the go to data source to detect flood events accurately. However, current methods are either based on fixed thresholds to differentiate water from land or train Artificial Intelligence (AI) models based on only S1 data, despite the availability of many other relevant data sources publicly. These models also lack comprehensive validations on out-of-sample data and deployment at scale. In this study, we investigated whether adding extra input layers could increase the performance of AI models in detecting floods from S1 data. We also provide performance across a range of 11 historical events, with results ranging between 0.93 and 0.97 accuracy, 0.53 and 0.81 IoU, and 0.68 and 0.89 F1 scores. Finally, we show the infrastructure we developed to deploy our AI models at scale to satisfy a range of use cases and user requests.

Flood risks have been increasing in recent years due to climate change. To provide flood risks through simulations near river areas, observational data of river discharges is required. However, the information of river discharges is sparse because of the limited availability of gauge stations. In this work, we use a combination of topographic features in location-scale with meteorological features in basin-scale to predict river discharges in ungauged locations along a river network. We utilize deep kernel learning as our core model and use non-stationary kernel formulation to enable spatiotemporal interpolation. We tested the performance in two different rivers using one year of collected data and compared it with several baselines. According to our experimental results, our proposed model improved Nash Efficiency Criterion (NSE) by 2.6%, normalized root mean-squared error (NRMSE) by 7%, and coefficient of determination (R2) by 275% compared with ordinary kriging as the baseline. The combination of features used in this experiment also improved NSE by 27.1%, NRMSE by 13.7%, and R2by 182.6%.

When providing the boundary conditions for hydrological flood models and estimating the associated risk, interpolating precipitation at very high temporal resolutions (e.g. 5 minutes) is essential not to miss the cause of flooding in local regions. In this paper, we study optical flow-based interpolation of globally available weather radar images from satellites. The proposed approach uses deep neural networks for the interpolation of multiple video frames, while terrain information is combined with temporarily coarse-grained precipitation radar observation as inputs for self-supervised training. An experiment with the Meteonet radar precipitation dataset for the flood risk simulation in Aude, a department in Southern France (2018), demonstrated the advantage of the proposed method over a linear interpolation baseline, with up to 20% error reduction.

For mitigating joint impact torques, researchers have reduced joint stiffness by series elastic actuators, reflected inertia by low gear ratios, and friction torque from drive-trains. However, these impact mitigation methods may impair the control performance of contact forces or may increase motor and robot mass. This paper proposes a design method for achieving a balance between impact mitigation performance and force control fidelity. We introduce an inertia-to-square-torque ratio as a new index for integrating the parameters of torque generation (motor continuous torque limits, gear ratios, etc.) and the parameters of impact mitigation (joint stiffness, reflected inertia, etc.). In the process, we make a hypothesis that a motor mass is negatively correlated with the ratio. Based on the hypothesis, we calculate a joint breakdown region of impact torques, joint stiffnesses, and motor masses. Finally, we decide the drive-train specifications of JAXON3-P and demonstrate that the proposed method provides high impact mitigation and force control capabilities through several experiments including the jumping motion of 0.3 m COG height.

We present a torque controller for humanoid robot's arm and a method to execute heavy-load tasks under that controller. Torque control of arms can reduce the joint load when an impulsive force is applied to the robot's hand. This feature is important for robots that will work among many humans or obstacles because accidental collisions may happen in such situations. We also developed a static force filter for force sensors at the end-effectors. This filter is utilized for the compensation for the static reaction force during heavy-load tasks. A life-sized humanoid robot JAXON digs soil with a shovel and carries soil with a wheelbarrow using our proposed controller.

Humanoid robots are expected to conduct tasks on behalf of humans in places such as a disaster scattered environment. Although humanoid robots have potentials to walk on uneven ground unlike wheeled robots, it is difficult to reach a given destination without falling down based on only visual information. In this paper, to reach the destination safely, we propose the autonomous safe locomotion system applying vision and sole reaction force to the footstep planning. Considering force information in addition to visual information, the robot can plan a path avoiding unstable footholds. The planned path is safer than a path which is planned based on only visual information. In our system, the robot checks if the foothold is safe or not by the foothold ascertainment motion. In addition to that, the robot saves the results of the motion to the database with the foothold label given by the visual classifier. To judge foothold safety, stiffness of the foothold is estimated from the reaction force and stepping amount. We propose the system considering these requirements for safe locomotion for bipedal robots and show experimental results using a real bipedal robot CHIDORI.

For humanoid robots, unexpected collision can cause instability of robot balancing and damage to both robots and environment. This paper presents a reactive bipedal walking controller against swing foot collision for humanoid robots. This controller is composed of following three components: 1) Swing Foot Trajectory Regenerator, 2) Swing Foot Collision Detector, and 3) Swing Foot Impact Mitigation Controller. By regenerating swing foot trajectory depending on situations, humanoid robots can avoid falling down. However, although humanoid robots detect collision and regenerate a swing foot, collision impact can cause bad effects such as damage and posture rotation. Therefore, to mitigate strong impact, we propose Swing Foot Impact Mitigation Controller, which is composed of two controllers. The proposed method is validated through the experiments by actual humanoid robot CHIDORI. We confirm that CHIDORI can avoid falling down against collision in two situations: walking on the flat ground, and stepping up a stair.

We propose a behavior learning method of a mobile robot using visual information. The proposed system consists of two networks for the state expression and the behavior generation. The state expression network, which is based on self-organizing algorithm, categorizes the robot's states. Also, the behavior generation network acquires the robot's behaviors using the output of the state expression network. Here, for training the behavior generation network, we utilize the actor-critic method as a kind of unsupervised learning scheme. We report some experimental results using the mobile robot in real environment. The mobile robot generates adaptive behaviors utilizing visual information.

We present a variety of vision-based adaptive and interactive behaviors in mechanical animals. The mechanical animal is a multi-legged robot designed as a remote-brained robots which does not bring its own brain within the body. It leaves the brain in the mother environment and talks with it by radio links. The brain is raised in the mother environment inherited over generations. The key idea of the remote-brained approach is that of interfacing intelligent software systems with real robot bodies through wireless technology. In this framework the robot system can have a powerful vision system in the brain environment. We have applied this approach toward formation of vision-based dynamic and intelligent behaviors of mechanical animals such as doglike robots and apelike robots. In this paper we introduce the remote-brained approach and describe some remote-brained robots and visual processes for adaptive and interactive behaviors with them.>