EP.65 A LEGGED ROBOT THAT PLAYS BADMINTON
Legged-wheeled delivery bots, Boston Dynamics presents Atlas adaptation & much more...
Adapting humanoid showcase from Boston Dynamics 🦾
Boston Dynamics has unveiled a new demonstration of its electric Atlas humanoid. The footage shows Atlas handling a complex manipulation task, identifying parts, understanding their geometry, and adjusting its strategy to successfully complete an assembly sequence.
One standout moment in the demo involves a deliberate disruption: an engineer drops an engine cover onto the floor mid-task. Atlas detects the anomaly, locates the object using its vision system, and re-engages with the task—evaluating the form of the dropped part and placing it back with precise alignment.
The robot’s adaptability relies on a sophisticated object pose estimation pipeline. The system uses monocular images and a render-and-compare approach trained on synthetic data. Given a CAD model and a rough pose estimate, Atlas refines the match by minimizing visual discrepancies, generating multiple pose hypotheses and selecting the best fit. This enables the robot to work reliably with novel parts, without manual tuning or retraining.
Beyond part recognition, Atlas demonstrates the ability to operate in dynamic environments while maintaining task objectives. The robot continually updates its internal world model to reflect changes in its surroundings, allowing it to stay on mission even in chaotic or unpredictable conditions.
Cobots as a tool to produce more engineers! 👷🏼
To prepare students for careers shaped by Industry 4.0, ABB has deployed 36 GoFa collaborative robots to 30 technical high schools across Slovenia. Backed by the country’s Ministry of Education and co-financed through the EU’s Next Generation initiative, the program introduces advanced robotics into classrooms, equipping students with practical automation skills.
Each school received ABB’s GoFa cobot along with access to RobotStudio, ABB’s simulation and programming environment. Students now gain hands-on experience in tasks like pick and place, inspection, and 3D printing—mirroring the workflows found in real-world manufacturing settings.
GoFa’s design emphasizes safety and ease of use, making the technology accessible to students with no prior robotics background. Through virtual simulation and real-world deployment, learners engage with robotics in a full-cycle engineering process. This exposure not only builds technical competence but also fosters creativity, problem-solving, and confidence. Educators report a noticeable uptick in student motivation and interest in automation-focused careers.
🦾 Feature sponsorship with ABB Robotics
A robot dog that plays badminton! 🏸
A research team from ETH Zurich has developed a unified control approach that enables legged mobile manipulators to play badminton, tackling one of the most complex coordination challenges in robotics, synchronizing perception, locomotion, and manipulation in fast-paced, dynamic environments.
The system uses a reinforcement learning-based whole-body control policy that manages all degrees of freedom to precisely track and strike a shuttlecock. A key innovation lies in the integration of a perception noise model, built from real-world camera data, which helps align the robot's sensory understanding in simulation with that in deployment. This model encourages the robot to learn active perception behaviors, such as adjusting head orientation during movement.
To ensure the robot can swing effectively and maintain control, the researchers also incorporated shuttlecock prediction, constrained reinforcement learning for motion stability, and system identification tools to prepare for real-world conditions. The result is a robot that can track shuttlecock trajectories, move efficiently across a service area, and return shots with precision—even against human opponents.
Meme of the week 🤖
Legged-wheeled robots deliver parcels in Austin! 📦
Veho Tech and Swiss robotics company RIVR have kicked off a new pilot program in Austin, deploying legged-wheeled robots for last-mile parcel delivery. This marks RIVR’s US debut, bringing its hybrid mobility systems to real-world e-commerce logistics.
Unlike sidewalk-bound delivery bots, RIVR’s robots are designed to handle urban complexity, navigating stairs, porches, and uneven ground. In this pilot, robots operate alongside Veho’s human couriers, handling up to 200 packages daily. Each unit is equipped with a secure cargo module and is remotely monitored during active delivery.
The robots are not replacing drivers. Instead, they complement them, maximizing efficiency in high-density areas where curb access and parking are often problematic. While a courier completes one delivery, a RIVR robot can independently complete another, including secure drop-off and photographic proof, integrated into Veho’s customer-facing app.
Backed by Veho’s infrastructure and relationships with major retailers like Saks Fifth Avenue and Nespresso, the trial aims to test scalability and user experience. RIVR, formerly Swiss Mile, positions its solution as a breakthrough in overcoming the “last-100-yard” challenge of urban delivery, going beyond what traditional wheeled or drone systems can achieve.
Light-powered ring robot that can crawl! 💍
Researchers at North Carolina State University have unveiled a light-powered soft robot that climbs thin wires and threads, mimicking the function of a cable car on a miniature scale. The device is made from a looped ribbon of liquid crystal elastomer twisted into a ring shape, resembling a rotini pasta coil.
When suspended on a thread or wire, either horizontal or inclined, the robot latches onto the track through two or three of its twisted loops, hanging beneath it. Upon exposure to overhead infrared light, the top section of the elastomer contracts and twists, acting like an auger to pull the robot forward along the track.
This motion is self-sustaining: the part exposed to light continuously shifts away from the source, while unexposed elastomer cycles into position, allowing for uninterrupted movement. The mechanism enables the robot to move autonomously as long as the light persists.
Lab tests demonstrated the robot’s versatility across tracks ranging from the width of a human hair to that of a drinking straw. It could climb slopes up to 80 degrees, traverse obstacles like knots, and carry payloads more than 12 times its own weight.
