Robotics News
The latest Robotics coverage curated for game developers.
arXiv cs.GR details InterMASH, a unified grasp-synthesis representation that packs hand, object, and contact data into sphere-fixed anchors. The goal is to make cross-embodiment training easier for human and robotic …
A new reconfigurable rover uses a variable-geometry truss to change its contact shape on the fly, cutting the coordination burden of traversal planning. The system reuses low-dimensional contact primitives instead …
A new humanoid retargeting system uses learned point-cloud correspondence to map human motion onto robots without hand-built keypoint pairs. That matters for teams training embodied agents: it promises denser pose …
A new reconstruction method turns curved capacitive textiles into a 3D shape sensor, letting robots infer nearby geometry from proximity fields instead of flat RGB or depth inputs. Proximity3D aggregates …
USC ICT and the US Army Research Laboratory built a dialogue-driven robot exploration workflow for low-signal, high-risk interiors. The system pairs spoken questions with sparse map updates and stills, helping …
A robotics team has built a bathing-demo pipeline that captures motion, body shape, contact regions, and force in sync, then transfers that data to a soft robotic hand. The system …
A new real-to-sim pipeline, R2S-EGO, uses dual proxies to patch sparse real-world capture with simulator-generated views. In tests on three Replica scenes and 48 frozen Unitree G1 ego views, six-view …
DiffPhysCam brings differentiable, physics-based camera simulation to inverse rendering and embodied AI. It gives developers finer control over camera intrinsics, optical artifacts like defocus blur, and calibration against real data. …
A new data-driven motion model can reproduce an individual's repetitive movement signature from real human data. Built around LSTM networks, it captures both velocity distribution and amplitude envelopes, which could …
AMD is pushing a full ROCm-based stack for embodied AI, showing that vision-language-action manipulation can run without CUDA lock-in. The demos span sim-to-real robot control, language-grounded object selection, Real2Sim data …
A new reconstruction method turns streaming point-cloud data into an orientable manifold mesh in real time, aiming to help underwater teams spot coverage gaps while they are still on site. …
NeuralActuator models servo behavior beyond the usual current-to-torque shortcut, aiming to cut sim-to-real drift on low-cost robots. It also infers external forces without dedicated sensors and flags mechanically restricted motion. …
A new humanoid control method uses the wrist as the handoff point between body motion and object manipulation. WristMimic tracks the torso and wrist kinematically, then lets the fingers learn …
A new octree update method called ECO keeps point-cloud mapping centered on a robot’s immediate surroundings instead of rebuilding a global tree every frame. On KITTI, it cut update time …
A new gradient-projection method for robotic multi-axis additive manufacturing keeps nozzle placement under 10 μm while cutting max joint jerk by up to 77.6%. For teams working on fabrication pipelines, …
Researchers synthesized 48,000 vision-language-kinematics trajectories from reconstructed indoor scenes to train a humanoid policy without any human demonstration data. The interesting part for developers is the sim-to-real pipeline: 3D Gaussian …
A new attention formulation treats each token as a matrix Lie group element instead of a feature vector, which is a neat fit for pose-heavy problems in graphics and robotics. …
Niantic says scans from millions of Pokémon Go players helped train a geospatial AI model now aimed at delivery robots and possibly military drones. For devs, the interesting part is …
TacCoRL reports a 72.5% average success rate on four contact-heavy bimanual tasks by adding tactile feedback to VLA policies and training in simulation. The practical takeaway is that the model …
A new navigation stack combines TSDF collision geometry with Gaussian Splatting’s visual richness, then adds YOLO-based object detection and 3D semantic lifting for planning. The practical win is safer, more …