Forget self-driving cars, robot driver Musashi takes on Tokyo roads

The concept of driverless cars has generated considerable excitement in recent times. Large companies have devoted a significant share of their research and development efforts to developing such autonomous driving technologies.

Researchers at the University of Tokyo have proposed a new, simpler, and cost-effective approach to the concept: giving humanoid robots the reins of cars for autonomous operations.

The team used a musculoskeletal humanoid robot named Musashi, designed to mimic the human body closely. The robot is expected to be capable of sitting in a car seat and operating the acceleration and brake pedals.

According to researchers, the robot’s flexibility, variable stiffness structure, and multiple sensors enabled it to perform steering wheel operations with both arms and recognize humans in the side mirror.

The details of the team’s research were published in the journal arXiv.

Human-like robots

Various studies on autonomous driving are underway to make safe and comfortable transportation a reality. Corporations have already accomplished headway vehicle following and autonomous parking, utilizing powerful cameras, LiDARs, GPS, and processors.

Since the DARPA Robotics Challenge (DRC), research has expanded to autonomous driving by humanoid robots equipped with sensors for visual, acoustic, and force information.

However, traditional humanoids seriously lack flexibility and proportionality, often requiring special kinds of equipment for particular tasks.

A solution is musculoskeletal humanoids that mimic human muscles with pneumatic actuators or muscle motors. Due to their flexibility and underactuation, they are said to excel at complex motions, including easily sitting in car seats and operating steering wheels with both arms.

As they possess a human-like structure, these robots are also suited for use as realistic crash test dummies.

Advanced robotic driver

Developed in 2019 by the research group, Musashi is a “musculoskeletal humanoid” designed as a testbed for learning control systems. Its form closely mirrors that of a human, incorporating a “joint and muscle structure” inspired by the human body’s proportions and mechanics.

The humanoid has 74 muscles and 39 joints, excluding the hand. The musculoskeletal structure comprises joint modules, muscle modules, and abrasion-resistant Dyneema fibers.

According to researchers, its muscles are arranged antagonistically around joints. Musashi’s modular body includes detailed joint and muscle modules with integrated sensors. The nonlinear elastic unit (NEU) uses rubber for flexibility.

Each of its movable eyes contains a high-resolution vision camera, capable of panning and tilting to capture various views, from straight ahead to quick side mirror checks. Its five-digit hands, located at the ends of jointed arms, are adept at turning the steering wheel based on the learning software and sensor data.

Musashi’s hands manage the handbrake, ignition key, and turn signals while its feet control the brake and accelerator pedals. Both extremities feature advanced sensors for measuring force and flexibility. The team installed a Wi-Fi router and Intel NUC PC in the vehicle to operate the recognition module and servo power supply.

A sophisticated version

During testing in a Toyota COMS electric car, Musashi completed a two-minute turn at a moderate speed by releasing the brake pedal without using the accelerator. It successfully navigated a junction by identifying traffic lights.

In another trial, Musashi attempted to use the accelerator but struggled to maintain consistent speed on hills. Apart from driving, the robot underwent tests to detect and react to objects in its path, including humans.

In spite of these obstacles, the researchers intend to create a more sophisticated version of Musashi with the goal of making further advancements.

“For autonomous driving by humanoids in the future, we would like to develop the next hardware and software using the obtained knowledge,” said researchers in the study.