IN Brief:
- Actus combines two Atlas M8500 RISC-V processors with dedicated motor-control acceleration and 512-bit vector processing.
- The proposed architecture includes 100-picosecond-class PWM, sigma-delta ADCs, three-port Time-Sensitive Networking, and hardware security functions.
- MIPS supports software-first development through a virtual platform, with Actus-1 evaluation hardware intended for subsequent prototype work.
MIPS has expanded its Actus robotic control platform, combining real-time RISC-V processing, motor-control acceleration, local inference, and deterministic networking within an architecture intended for compact robotic actuators.
The design integrates two Atlas M8500 processors with dedicated control hardware, a 512-bit vector-processing capability, and communications interfaces supporting coordinated operation across multiple robot joints. MIPS is developing the architecture around GlobalFoundries' FDX fully depleted silicon-on-insulator technology.
The latest development targets the electronic control functions required where a robot's mechanical movement is generated. In a multi-joint assembly, each actuator must respond to movement commands while maintaining suitable control of motor current, torque, position, and communication with the wider system.
These functions place different demands on processing hardware. A high-level controller may calculate the movement required to reach a destination, while the electronics inside an individual joint must update switching commands and evaluate feedback at predictable intervals.
Delays in the local control process can alter the response of a motor, particularly where several actuators must coordinate their movement. Processor throughput alone does not establish whether every time-critical operation will complete within its required deadline.
Actus uses two MIPS Atlas M8500 real-time RISC-V processors with hardware multithreading. MIPS states that the architecture removes context-switch penalties during interrupt servicing, reducing some of the execution overhead associated with responding to time-sensitive events.
Hardware multithreading allows processing resources to manage more than one execution context without relying entirely on conventional software task switching. The resulting control performance still depends on interrupt behaviour, workload allocation, and the timing requirements of the application.
The platform also incorporates dedicated acceleration for trigonometric calculations used in field-oriented motor control. MIPS reports performance exceeding that of standard trigonometric libraries by more than 16 times for the relevant operations.
Field-oriented control regulates the current supplied to an electric motor by representing measured phase currents in a coordinate system aligned with the motor's magnetic field. The mathematical transformations allow the control algorithm to distinguish between current components associated with magnetic flux and torque production.
These calculations must be repeated as the motor operates, requiring the controller to process electrical measurements and update its switching outputs within the available control interval. Accelerating the mathematics can reduce the processing time devoted to those operations.
The manufacturer's performance comparison concerns specified trigonometric calculations, not the execution speed of an entire motor-control application. Actual control-loop performance also depends on sensor acquisition, processor scheduling, memory access, and the time required to produce the switching commands.
Actus includes pulse-width modulation hardware with a stated resolution in the 100-picosecond class. PWM outputs regulate the switching devices used in motor power stages, allowing the controller to adjust the electrical power delivered to the motor windings.
Fine timing resolution provides scope for adjusting switching events in small increments. However, the precision of the resulting movement depends on the motor, power electronics, feedback sensors, mechanical transmission, and control algorithm, rather than the PWM resolution alone.
Sigma-delta analogue-to-digital conversion provides a route for acquiring electrical measurements used by the motor controller. Current feedback is particularly important because the control algorithm must respond to the electrical state of the motor as well as to the requested movement.
Measurement quality depends on the sensing arrangement and its interaction with switching noise. Converter resolution, sampling conditions, electrical filtering, and signal timing all influence the information available to the processor.
Actus extends beyond motor control through a 512-bit ARC VPX vector-processing capability. MIPS intends this resource to support functions such as sensor fusion and predictive maintenance alongside the time-critical tasks involved in operating the actuator.
Sensor fusion combines information from several measurements to estimate the state of a system. Within a robot joint, electrical feedback, position information, and other sensor data can contribute to control or condition-monitoring functions.
Predictive maintenance algorithms can examine changes in measured behaviour that may indicate developing mechanical or electrical problems. Their effectiveness depends on the available data, the training or diagnostic method, and the extent to which ordinary operating variations can be distinguished from faults.
Executing these functions locally may reduce the amount of information that must be transmitted to a central processor. The controller must nevertheless prevent supporting workloads from interfering with time-critical motor-control operations.
Communication between joints is addressed through integrated Gigabit Ethernet with three-port Time-Sensitive Networking, alongside CAN FD and serial interfaces. TSN provides mechanisms for coordinating traffic and timing across compatible Ethernet equipment.
Deterministic networking can support robotic systems in which movement commands and feedback must be exchanged within defined intervals. Its effectiveness depends on compatible endpoints, network configuration, traffic scheduling, and the timing behaviour of the connected controllers.
The proposed architecture also includes a hardware security module supporting post-quantum cryptography. These capabilities are intended to help protect individual actuator nodes and their communications, although the security of a complete robotic system depends on its software, configuration, credentials, and network architecture.
MIPS identifies functional-safety support as another element of Actus. Such features can provide resources for developing safety-oriented equipment but do not establish that an individual actuator or completed robot has passed a particular functional-safety assessment.
The underlying semiconductor technology is GlobalFoundries' FDX platform, with MIPS identifying AutoPro150 and integrated MRAM as supporting capabilities. The architecture is intended to combine processing, control, communications, and memory functions within the limited space and thermal capacity available in robotic joints.
The commercial implementation remains subject to engineering development. MIPS is providing a virtual platform so that software teams can begin work before final silicon is available, with the Actus-1 evaluation board and SDK forming part of the route towards physical prototypes.
Actus also forms part of MIPS' wider physical AI portfolio, alongside the Acies inference architecture and Aegis safety-oriented computing platform. The three address different processing requirements, while Actus concentrates on translating computed decisions into controlled physical movement.
MIPS will demonstrate the platform at electronica 2026 in Munich from 10 to 13 November. The company has not announced a volume-production date for customer-specific Actus silicon or published complete system-level benchmarks establishing its performance in a production robotic actuator.
